Prepare AFM tip solvation¶
This notebook demonstrates deposition of an SDS adsorption layer on a non-spherical AFM tip model.
Initialization¶
IPython magic¶
In [1]:
%load_ext autoreload
%autoreload 3
Imports¶
In [2]:
import ase.io # here used for reading pdb files
from ase.visualize import view
from ase.visualize.plot import plot_atoms # has nasty offset issues
from cycler import cycler # here used for cycling through colors in plots
import datetime
import fabric # for pythonic ssh connections
from fireworks import LaunchPad, Firework, Tracker, Workflow
from fireworks import FileTransferTask, PyTask, ScriptTask
# FireWorks functionality
from fireworks import Firework, LaunchPad, ScriptTask, Workflow
from fireworks.user_objects.firetasks.templatewriter_task import TemplateWriterTask
from fireworks.user_objects.firetasks.filepad_tasks import AddFilesTask, GetFilesTask, GetFilesByQueryTask
from imteksimfw.fireworks.user_objects.firetasks.cmd_tasks import CmdTask
from fireworks.utilities.filepad import FilePad # direct FilePad access, similar to the familiar LaunchPad
import glob
import gc # manually clean up memory with gc.collect()
import gromacs # GromacsWrapper, here used for evoking gmc commands, reading and writing .ndx files
# from io import StringIO, TextIOWrapper
import io
import itertools # for products of iterables
import json # generic serialization of lists and dicts
import jinja2 # here used for filling packmol input script template
import jinja2.meta # for gathering variables in a jinja2 template
import logging
import matplotlib.pyplot as plt
import MDAnalysis as mda # here used for reading and analyzing gromacs trajectories
import MDAnalysis.analysis.rdf as mda_rdf
import MDAnalysis.analysis.rms as mda_rms
from mpl_toolkits.mplot3d import Axes3D # here used for 3d point cloud scatter plot
import miniball # finds minimum bounding sphere of a point set
import nglview
import numpy as np
import os, os.path
import pandas as pd
import panedr # reads GROMACS edr into pandas df, requires pandas and pbr
import parmed as pmd # has quite a few advantages over ASE when it comes to parsing pdb
from pprint import pprint
import pymongo # for sorting in queries
import scipy.constants as sc
import subprocess # used for evoking external packmol
import sys
import tempfile
import yaml
GromacsWrapper might need a file ~/.gromacswrapper.cfg with content
[Gromacs]
tools = gmx gmx_d
# gmx_mpi_d gmx_mpi_d
# name of the logfile that is written to the current directory
logfilename = gromacs.log
# loglevels (see Python's logging module for details)
# ERROR only fatal errors
# WARN only warnings
# INFO interesting messages
# DEBUG everything
# console messages written to screen
loglevel_console = INFO
# file messages written to logfilename
loglevel_file = DEBUG
in order to know the GROMACS executables it is allowed to use. Otherwise,
calls to gmx_mpi or gmx_mpi_d without MPI wrapper might lead to MPI
warnings in output that cause GromacsWrapper to fail.
Logging¶
In [3]:
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger()
logger.setLevel(logging.INFO)
ParmEd needs to know the GROMACS topology folder, usually get this from
envionment variable GMXLIB:
Function definitions¶
In [4]:
def find_undeclared_variables(infile):
"""identify all variables evaluated in a jinja 2 template file"""
env = jinja2.Environment()
with open(infile) as template_file:
parsed = env.parse(template_file.read())
undefined = jinja2.meta.find_undeclared_variables(parsed)
return undefined
In [5]:
def plot_side_views_with_spheres(atoms, cc, R, figsize=(12,4), fig=None, ax=None):
"""
Plots xy, yz and zx projections of atoms and sphere(s)
Parameters
----------
atoms: ase.atoms
cc: (N,3) ndarray
centers of spheres
R: (N,) ndarray
radii of spheres
figsize: 2-tuple, default (12,4)
fig: matplotlib.figure, default None
ax: list of three matploblib.axes objects
"""
logger = logging.getLogger(__name__)
atom_radii = 0.5
cc = np.array(cc,ndmin=2)
logger.info("C({}) = {}".format(cc.shape,cc))
R = np.array(R,ndmin=1)
logger.info("R({}) = {}".format(R.shape,R))
xmin = atoms.get_positions().min(axis=0)
xmax = atoms.get_positions().max(axis=0)
logger.info("xmin({}) = {}".format(xmin.shape,xmin))
logger.info("xmax({}) = {}".format(xmax.shape,xmax))
### necessary due to ASE-internal atom position computations
# see https://gitlab.com/ase/ase/blob/master/ase/io/utils.py#L69-82
X1 = xmin - atom_radii
X2 = xmax + atom_radii
M = (X1 + X2) / 2
S = 1.05 * (X2 - X1)
scale = 1
internal_offset = [ np.array(
[scale * np.roll(M,i)[0] - scale * np.roll(S,i)[0] / 2,
scale * np.roll(M,i)[1] - scale * np.roll(S,i)[1] / 2]) for i in range(3) ]
###
atom_permut = [ atoms.copy() for i in range(3) ]
for i, a in enumerate(atom_permut):
a.set_positions( np.roll(a.get_positions(),i,axis=1) )
rot = ['0x,0y,0z']*3#,('90z,90x'),('90x,90y,0z')]
label = [ np.roll(np.array(['x','y','z'],dtype=str),i)[0:2] for i in range(3) ]
# dim: sphere, view, coord
center = np.array([
[ np.roll(C,i)[0:2] - internal_offset[i] for i in range(3) ] for C in cc ])
logger.info("projected cc({}) = {}".format(center.shape,center))
color_cycle = cycler(color=[
'tab:blue', 'tab:orange', 'tab:green', 'tab:red', 'tab:purple'])
circle = [ [ plt.Circle( c , r, fill=False, **col) for c in C ] for C,r,col in zip(center,R,color_cycle) ]
margin = 1.1
# dim: view, coord, minmax (i.e., 3,2,2)
plot_bb = np.rollaxis( np.array(
[ np.min(center - margin*(np.ones(center.T.shape)*R).T,axis=0),
np.max(center + margin*(np.ones(center.T.shape)*R).T,axis=0) ] ).T, 1, 0)
#plot_bb = np.array( [ [
# [ [np.min(c[0]-margin*R[0]), np.max(c[0]+margin*R[0])],
# [np.min(c[1]-margin*R[0]), np.max(c[1]+margin*R[0])] ] ) for c in C ] for C,r in zip(center,R) ] )
logger.info("projected bb({}) = {}".format(plot_bb.shape,plot_bb))
if ax is None:
fig, ax = plt.subplots(1,3,figsize=figsize)
(ax_xy, ax_xz, ax_yz) = ax[:]
logger.info("iterators len(atom_permut={}, len(ax)={}, len(rot)={}, len(circle)={}".format(
len(atom_permut),len(ax),len(rot),len(circle)))
#logger.info("len(circle)={}".format(len(circle))
#for aa, a, r, C in zip(atom_permut,ax,rot,circle):
for i, a in enumerate(ax):
# rotation strings see https://gitlab.com/ase/ase/blob/master/ase/utils/__init__.py#L235-261
plot_atoms(atom_permut[i],a,rotation=rot[i],radii=0.5,show_unit_cell=0,offset=(0,0))
for j, c in enumerate(circle):
logger.info("len(circle[{}])={}".format(j,len(c)))
a.add_patch(c[i])
for a,l,bb in zip(ax,label,plot_bb):
a.set_xlabel(l[0])
a.set_ylabel(l[1])
a.set_xlim(*bb[0,:])
a.set_ylim(*bb[1,:])
return fig, ax
In [6]:
def pack_sphere(C,
R_inner_constraint, # shell inner radius
R_outer_constraint, # shell outer radius
sfN, # number of surfactant molecules
inner_atom_number, # inner atom
outer_atom_number, # outer atom
surfactant = 'SDS',
counterion = 'NA',
tolerance = 2):
"""Creates context for filling Jinja2 PACKMOL input template in order to
generate preassembled surfactant spheres with couinterions at polar heads"""
logger = logging.getLogger(__name__)
logger.info(
"sphere with {:d} surfactant molecules in total.".format(sfN ) )
# sbX, sbY, sbZ = sb_measures
# spheres parallelt to x-axis
sphere = {}
ionsphere = {}
# surfactant spheres
# inner constraint radius: R + 1*tolerance
# outer constraint radius: R + 1*tolerance + l_surfactant
# ions between cylindric planes at
# inner radius: R + 1*tolerance + l_surfactant
# outer radius: R + 2*tolerance + l_surfactant
sphere["surfactant"] = surfactant
sphere["inner_atom_number"] = inner_atom_number
sphere["outer_atom_number"] = outer_atom_number
sphere["N"] = sfN
sphere["c"] = C
sphere["r_inner"] = R_inner
sphere["r_inner_constraint"] = R_inner_constraint
sphere["r_outer_constraint"] = R_outer_constraint
sphere["r_outer"] = R_outer
logging.info(
"sphere with {:d} molecules at {}, radius {}".format(
sphere["N"], sphere["c"], sphere["r_outer"]))
# ions at outer surface
ionsphere["ion"] = counterion
ionsphere["N"] = sphere["N"]
ionsphere["c"] = sphere["c"]
ionsphere["r_inner"] = sphere["r_outer"]
ionsphere["r_outer"] = sphere["r_outer"] + tolerance
# experience shows: movebadrandom advantegous for (hemi-) spheres
context = {
'spheres': [sphere],
'ionspheres': [ionsphere],
'movebadrandom': True,
}
return context
In [7]:
def memuse():
"""Quick overview on memory usage of objects in Jupyter notebook"""
# https://stackoverflow.com/questions/40993626/list-memory-usage-in-ipython-and-jupyter
# These are the usual ipython objects, including this one you are creating
ipython_vars = ['In', 'Out', 'exit', 'quit', 'get_ipython', 'ipython_vars']
# Get a sorted list of the objects and their sizes
return sorted([(x, sys.getsizeof(globals().get(x))) for x in dir(sys.modules['__main__']) if not x.startswith('_') and x not in sys.modules and x not in ipython_vars], key=lambda x: x[1], reverse=True)
Global settings¶
In [8]:
os.environ['GMXLIB']
Out[8]:
In [9]:
pmd.gromacs.GROMACS_TOPDIR = os.environ['GMXLIB']
In [10]:
# prefix = '/mnt/dat/work/testuser/indenter/sandbox/20191110_packmol'
prefix = '/home/jotelha/git/N_surfactant_on_substrate_template'
In [11]:
work_prefix = '/home/jotelha/tmp/20200329_fw/'
In [11]:
os.chdir(work_prefix)
HPC-related settings¶
In [ ]:
hpc_max_specs = {
'forhlr2': {
'fw_queue_category': 'forhlr2_queue',
'fw_noqueue_category': 'forhlr2_noqueue',
'queue':'develop',
'physical_cores_per_node': 20,
'logical_cores_per_node': 40,
'nodes': 4,
'walltime': '00:60:00'
},
'juwels_devel': {
'fw_queue_category': 'juwels_queue',
'fw_noqueue_category': 'juwels_noqueue',
'queue':'devel',
'physical_cores_per_node': 48,
'logical_cores_per_node': 96,
'nodes': 8,
'walltime': '00:30:00'
},
'juwels': {
'fw_queue_category': 'juwels_queue',
'fw_noqueue_category': 'juwels_noqueue',
'queue':'batch',
'physical_cores_per_node': 48,
'logical_cores_per_node': 96,
'nodes': 1024,
'walltime': '00:30:00'
}
}
In [ ]:
std_exports = {
'forhlr2': {
'OMP_NUM_THREADS': 1,
'KMP_AFFINITY': "'verbose,compact,1,0'",
'I_MPI_PIN_DOMAIN':'core'
},
'juwels': {
'OMP_NUM_THREADS': 1,
'KMP_AFFINITY': "'verbose,compact,1,0'",
'I_MPI_PIN_DOMAIN':'core'
}
}
FireWorks LaunchPad and FilePad¶
In [ ]:
# the FireWorks LaunchPad
lp = LaunchPad.auto_load() #Define the server and database
In [ ]:
# FilePad behaves analogous to LaunchPad
fp = FilePad.auto_load()
Sub-WF: PACKMOL¶
In [27]:
def sub_wf_pack(d, fws_root):
global project_id, \
C, R_inner_constraint, R_outer_constraint, \
tail_atom_number, head_atom_number, surfactant, counterion, tolerance, \
fw_name_template, hpc_max_specs, machine
# TODO: instead of global variables, use class
fw_list = []
### Template
files_in = {'input_file': 'input.template' }
files_out = { 'input_file': 'input.inp' }
# exports = std_exports[machine].copy()
# Jinja2 context:
packmol_script_context = {
'header': '{:s} packing SDS around AFM probe model'.format(project_id),
'system_name': '{:d}_SDS_on_50_Ang_AFM_tip_model'.format(d["nmolecules"]),
'tolerance': tolerance,
'write_restart': True,
'static_components': [
{
'name': 'indenter'
}
]
}
# use pack_sphere function at the notebook's head to generate template context
packmol_script_context.update(
pack_sphere(
C,R_inner_constraint,R_outer_constraint, d["nmolecules"],
tail_atom_number+1, head_atom_number+1, surfactant, counterion, tolerance))
ft_template = TemplateWriterTask( {
'context': packmol_script_context,
'template_file': 'input.template',
'template_dir': '.',
'output_file': 'input.inp'} )
fw_template = Firework([ft_template],
name = ', '.join(('template', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'fill_template',
**d
}
},
parents = fws_root )
fw_list.append(fw_template)
### PACKMOL
files_in = {
'input_file': 'input.inp',
'indenter_file': 'indenter.pdb',
'surfatcant_file': '1_SDS.pdb',
'counterion_file': '1_NA.pdb' }
files_out = {
'data_file': '*_packmol.pdb'}
ft_pack = CmdTask(
cmd='packmol',
opt=['< input.inp'],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True)
fw_pack = Firework([ft_pack],
name = ', '.join(('packmol', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_queue_category'],
'_queueadapter': {
'queue': hpc_max_specs[machine]['queue'],
'walltime' : hpc_max_specs[machine]['walltime'],
'ntasks': 1,
},
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'packmol',
**d
}
},
parents = [ *fws_root, fw_template] )
fw_list.append(fw_pack)
return fw_list, fw_pack
def sub_wf_pack_push(d, fws_root):
global project_id, hpc_max_specs, machine
fw_list = []
files_in = {'data_file': 'packed.pdb' }
fts_push = [ AddFilesTask( {
'compress': True ,
'paths': "packed.pdb",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'initial_config',
**d }
} ) ]
fw_push = Firework(fts_push,
name = ', '.join(('transfer', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'transfer',
**d
}
},
parents = fws_root )
fw_list.append(fw_push)
return fw_list, fw_push
Sub-WF: GMX prep¶
In [28]:
def sub_wf_gmx_prep_pull(d, fws_root):
global project_id, hpc_max_specs, machine
fw_list = []
files_in = {}
files_out = { 'data_file': 'in.pdb' }
fts_pull = [ GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'initial_config',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['in.pdb'] ) ]
fw_pull = Firework(fts_pull,
name = ', '.join(('fetch', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'fetch',
**d
}
},
parents = fws_root )
fw_list.append(fw_pull)
return fw_list, fw_pull
def sub_wf_gmx_prep(d, fws_root):
global project_id, hpc_max_specs, machine
# TODO: instead of global variables, use class
fw_list = []
### PDB chain
files_in = {'data_file': 'in.pdb' }
files_out = {'data_file': 'out.pdb'}
fts_pdb_chain = CmdTask(
cmd='pdb_chain',
opt=['< in.pdb > out.pdb'],
store_stdout = False,
store_stderr = False,
use_shell = True,
fizzle_bad_rc= True)
fw_pdb_chain = Firework(fts_pdb_chain,
name = ', '.join(('pdb_chain', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'pdb_chain',
**d
}
},
parents = fws_root )
fw_list.append(fw_pdb_chain)
### PDB tidy
files_in = {'data_file': 'in.pdb' }
files_out = {'data_file': 'out.pdb'}
fts_pdb_tidy = CmdTask(
cmd='pdb_tidy',
opt=['< in.pdb > out.pdb'],
store_stdout = False,
store_stderr = False,
use_shell = True,
fizzle_bad_rc= True)
fw_pdb_tidy = Firework(fts_pdb_tidy,
name = ', '.join(('pdb_tidy', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'pdb_tidy',
**d
}
},
parents = [ fw_pdb_chain ] )
fw_list.append(fw_pdb_tidy)
### GMX pdb2gro
files_in = {'data_file': 'in.pdb' }
files_out = {
'data_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp'}
fts_gmx_pdb2gro = [ CmdTask(
cmd='gmx',
opt=['pdb2gmx',
'-f', 'in.pdb',
'-o', 'default.gro',
'-p', 'default.top',
'-i', 'default.posre.itp',
'-ff', 'charmm36',
'-water' , 'tip3p'],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True) ]
fw_gmx_pdb2gro = Firework(fts_gmx_pdb2gro,
name = ', '.join(('gmx_pdb2gro', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_pdb2gro',
**d
}
},
parents = [ fw_pdb_tidy ] )
fw_list.append(fw_gmx_pdb2gro)
### GMX editconf
files_in = {
'data_file': 'in.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp'}
files_out = {
'data_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp'}
fts_gmx_editconf = [ CmdTask(
cmd='gmx',
opt=['editconf',
'-f', 'in.gro',
'-o', 'default.gro',
'-d', 2.0, # distance between content and box boundary in nm
'-bt', 'cubic', # box type
],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True) ]
fw_gmx_editconf = Firework(fts_gmx_editconf,
name = ', '.join(('gmx_editconf', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_editconf',
**d
}
},
parents = [ fw_gmx_pdb2gro ] )
fw_list.append(fw_gmx_editconf)
return fw_list, fw_gmx_editconf
def sub_wf_gmx_prep_push(d, fws_root):
global project_id, hpc_max_specs, machine
fw_list = []
files_in = {
'data_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp' }
fts_push = [
AddFilesTask( {
'compress': True ,
'paths': "default.gro",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'initial_config_gro',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.top",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'initial_config_top',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.posre.itp",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'initial_config_posre_itp',
**d }
} ) ]
fw_push = Firework(fts_push,
name = ', '.join(('push', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'push',
**d
}
},
parents = fws_root )
fw_list.append(fw_push)
return fw_list, fw_push
Sub-WF: GMX EM¶
In [221]:
def sub_wf_gmx_em_pull(d, fws_root):
global project_id, source_project_id, hpc_max_specs, machine
fw_list = []
files_in = {}
files_out = {
'data_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp',
}
fts_fetch = [
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'initial_config_gro',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.gro'] ),
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'initial_config_top',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.top'] ),
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'initial_config_posre_itp',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.posre.itp'] ) ]
fw_fetch = Firework(fts_fetch,
name = ', '.join(('fetch', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'fetch',
**d
}
},
parents = fws_root )
fw_list.append(fw_fetch)
return fw_list, fw_fetch
def sub_wf_gmx_em(d, fws_root):
global project_id, hpc_max_specs, machine
fw_list = []
### GMX grompp
files_in = {
'input_file': 'default.mdp',
'data_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp'}
files_out = {
'input_file': 'default.tpr',
'parameter_file': 'mdout.mdp' }
fts_gmx_grompp = [ CmdTask(
cmd='gmx',
opt=['grompp',
'-f', 'default.mdp',
'-c', 'default.gro',
'-r', 'default.gro',
'-o', 'default.tpr',
'-p', 'default.top',
],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True) ]
fw_gmx_grompp = Firework(fts_gmx_grompp,
name = ', '.join(('gmx_grompp_em', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_grompp_em',
**d
}
},
parents = fws_root )
fw_list.append(fw_gmx_grompp)
### GMX mdrun
files_in = {'input_file': 'em.tpr'}
files_out = {
'log_file': 'em.log',
'energy_file': 'em.edr',
'trajectory_file': 'em.trr',
'data_file': 'em.gro' }
fts_gmx_mdrun = [ CmdTask(
cmd='gmx',
opt=[' mdrun',
'-deffnm', 'em', '-v' ],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True) ]
fw_gmx_mdrun = Firework(fts_gmx_mdrun,
name = ', '.join(('gmx_mdrun_em', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_queue_category'],
'_queueadapter': {
'queue': hpc_max_specs[machine]['queue'],
'walltime' : hpc_max_specs[machine]['walltime'],
'ntasks': 96,
},
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_mdrun_em',
**d
}
},
parents = [ fw_gmx_grompp ] )
fw_list.append(fw_gmx_mdrun)
return fw_list, fw_gmx_mdrun
def sub_wf_gmx_em_push(d, fws_root):
global project_id, hpc_max_specs, machine
fw_list = []
files_in = {
'log_file': 'em.log',
'energy_file': 'em.edr',
'trajectory_file': 'em.trr',
'data_file': 'em.gro' }
files_out = {}
fts_push = [
AddFilesTask( {
'compress': True ,
'paths': "em.log",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'em_log',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "em.edr",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'em_edr',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "em.trr",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'em_trr',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "em.gro",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'em_gro',
**d }
} ) ]
fw_push = Firework(fts_push,
name = ', '.join(('push', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'push',
**d
}
},
parents = fws_root )
fw_list.append(fw_push)
return fw_list, fw_push
Sub-WF: pulling preparations¶
In [235]:
def sub_wf_pull_prep_pull(d, fws_root):
global project_id, source_project_id, hpc_max_specs, machine
fw_list = []
files_in = {}
files_out = {
'data_file': 'default.gro',
}
fts_pull = [
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'initial_config_gro',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.gro'] ) ]
fw_pull = Firework(fts_pull,
name = ', '.join(('fetch', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'fetch',
**d
}
},
parents = fws_root )
fw_list.append(fw_pull)
return fw_list, fw_pull
def sub_wf_pull_prep(d, fws_root):
global project_id, \
surfactant, counterion, substrate, nsubstrate, \
tail_atom_name, \
fw_name_template, hpc_max_specs, machine
fw_list = []
### System topolog template
files_in = { 'template_file': 'sys.top.template' }
files_out = { 'topology_file': 'sys.top' }
# Jinja2 context:
template_context = {
'system_name': '{nsurfactant:d}_{surfactant:s}_{ncounterion:d}_{counterion:s}_{nsubstrate:d}_{substrate:s}'.format(
project_id=project_id,
nsurfactant=d["nmolecules"], surfactant=surfactant,
ncounterion=d["nmolecules"], counterion=counterion,
nsubstrate=nsubstrate, substrate=substrate),
'header': '{project_id:s}: {nsurfactant:d} {surfactant:s} and {ncounterion:d} {counterion:s} around {nsubstrate:d}_{substrate:s} AFM probe model'.format(
project_id=project_id,
nsurfactant=d["nmolecules"], surfactant=surfactant,
ncounterion=d["nmolecules"], counterion=counterion,
nsubstrate=nsubstrate, substrate=substrate),
'nsurfactant': d["nmolecules"],
'surfactant': surfactant,
'ncounterion': d["nmolecules"],
'counterion': counterion,
'nsubstrate': nsubstrate,
'substrate': substrate,
}
fts_template = [ TemplateWriterTask( {
'context': template_context,
'template_file': 'sys.top.template',
'template_dir': '.',
'output_file': 'sys.top'} ) ]
fw_template = Firework(fts_template,
name = ', '.join(('template', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'fill_template',
**d
}
},
parents = fws_root )
fw_list.append(fw_template)
### Index file
files_in = {'data_file': 'default.gro'}
files_out = {'index_file': 'default.ndx'}
fts_gmx_make_ndx = [
CmdTask(
cmd='gmx',
opt=['make_ndx',
'-f', 'default.gro',
'-o', 'default.ndx',
],
env = 'python',
stdin_key = 'stdin',
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
fizzle_bad_rc= True) ]
fw_gmx_make_ndx = Firework(fts_gmx_make_ndx,
name = ', '.join(('gmx_make_ndx', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'stdin': 'q\n',
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_make_ndx',
**d
}
},
parents = fws_root )
fw_list.append(fw_gmx_make_ndx)
### pulling groups
files_in = {
'data_file': 'default.gro',
'topology_file': 'default.top',
'index_file': 'in.ndx',
'parameter_file': 'in.mdp',
}
files_out = {
'data_file': 'default.gro', # pass through unmodified
'topology_file': 'default.top', # pass unmodified
'index_file': 'out.ndx',
'input_file': 'out.mdp',
}
fts_make_pull_groups = [ CmdTask(
cmd='gmx_tools',
opt=['--verbose', '--log', 'default.log',
'make','pull_groups',
'--topology-file', 'default.top',
'--coordinates-file', 'default.gro',
'--residue-name', surfactant,
'--atom-name', tail_atom_name,
'--reference-group-name', 'Substrate',
'-k', 1000,
'--rate', 0.1, '--',
'in.ndx', 'out.ndx', 'in.mdp', 'out.mdp'],
stderr_file = 'std.err',
stdout_file = 'std.out',
env = 'python',
store_stdout = True,
store_stderr = True,
fizzle_bad_rc= True) ]
fw_make_pull_groups = Firework(fts_make_pull_groups,
name = ', '.join(('gmx_tools_make_pull_groups', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_tools_make_pull_groups',
**d
}
},
parents = [ *fws_root, fw_template, fw_gmx_make_ndx] )
fw_list.append(fw_make_pull_groups)
return fw_list, fw_make_pull_groups
def sub_wf_pull_prep_push(d, fws_root):
global project_id, hpc_max_specs, machine
fw_list = []
files_in = {
'topology_file': 'default.top',
'index_file': 'default.ndx',
'input_file': 'default.mdp',
}
fts_push = [
AddFilesTask( {
'compress': True ,
'paths': "default.top",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'top_pull',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.ndx",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'ndx_pull',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.mdp",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'mdp_pull',
**d }
} ),
]
fw_push = Firework(fts_push,
name = ', '.join(('push', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'push',
**d
}
},
parents = fws_root )
fw_list.append(fw_push)
return fw_list, fw_push
Sub-WF: GMX pull¶
In [271]:
def sub_wf_gmx_pull_pull(d, fws_root):
global project_id, source_project_id, hpc_max_specs, machine
fw_list = []
files_in = {}
files_out = {
'data_file': 'default.gro',
'topology_file': 'default.top',
'input_file': 'default.mdp',
'indef_file': 'default.ndx',
}
fts_fetch = [
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id, # earlier
'metadata->type': 'initial_config_gro',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.gro'] ),
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'top_pull',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.top'] ),
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'ndx_pull',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.ndx'] ),
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'mdp_pull',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.mdp'] ) ]
fw_fetch = Firework(fts_fetch,
name = ', '.join(('pull', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'pull',
**d
}
},
parents = fws_root )
fw_list.append(fw_fetch)
return fw_list, fw_fetch
def sub_wf_gmx_pull(d, fws_root):
global project_id, hpc_max_specs, machine
fw_list = []
### GMX grompp
files_in = {
'input_file': 'default.mdp',
'index_file': 'default.ndx',
'data_file': 'default.gro',
'topology_file': 'default.top'}
files_out = {
'input_file': 'default.tpr',
'parameter_file': 'mdout.mdp',
}
# gmx grompp -f pull.mdp -n pull_groups.ndx -c em.gro -r em.gro -o pull.tpr -p sys.top
fts_gmx_grompp = [ CmdTask(
cmd='gmx',
opt=['grompp',
'-f', 'default.mdp', # parameter file
'-n', 'default.ndx', # index file
'-c', 'default.gro', # coordinates file
'-r', 'default.gro', # restraint positions
'-p', 'default.top', # topology file
'-o', 'default.tpr', # compiled output
],
env = 'python',
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
fizzle_bad_rc= True) ]
fw_gmx_grompp = Firework(fts_gmx_grompp,
name = ', '.join(('gmx_grompp_pull', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_grompp_pull',
**d
}
},
parents = fws_root )
fw_list.append(fw_gmx_grompp)
### GMX mdrun
files_in = {'input_file': 'default.tpr'}
files_out = {
'log_file': 'default.log',
'energy_file': 'default.edr',
'trajectory_file': 'default.trr',
'compressed_trajectory_file': 'default.xtc',
'data_file': 'default.gro',
'pullf_file': 'default_pullf.xvg',
'pullx_file': 'default_pullx.xvg',}
fts_gmx_mdrun = [ CmdTask(
cmd='gmx',
opt=['mdrun',
'-deffnm', 'default', '-v' ],
env='python',
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
fizzle_bad_rc= True) ]
fw_gmx_mdrun = Firework(fts_gmx_mdrun,
name = ', '.join(('gmx_mdrun_pull', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_queue_category'],
'_queueadapter': {
'queue': hpc_max_specs[machine]['queue'],
'walltime' : hpc_max_specs[machine]['walltime'],
'ntasks': hpc_max_specs[machine]['physical_cores_per_node'],
'ntasks_per_node': hpc_max_specs[machine]['physical_cores_per_node'],
# JUWELS GROMACS
# module("load","Stages/2019a","Intel/2019.3.199-GCC-8.3.0","IntelMPI/2019.3.199")
# module("load","GROMACS/2019.3","GROMACS-Top/2019.3")
# fails with segmentation fault when using SMT (96 logical cores)
},
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_mdrun_pull',
**d
}
},
parents = [ fw_gmx_grompp ] )
fw_list.append(fw_gmx_mdrun)
return fw_list, fw_gmx_mdrun
def sub_wf_gmx_pull_push(d, fws_root):
global project_id, hpc_max_specs, machine
fw_list = []
files_in = {
'log_file': 'default.log',
'energy_file': 'default.edr',
'trajectory_file': 'default.trr',
'compressed_trajectory_file': 'default.xtc',
'data_file': 'default.gro',
'pullf_file': 'default_pullf.xvg',
'pullx_file': 'default_pullx.xvg',}
files_out = {}
fts_push = [
AddFilesTask( {
'compress': True ,
'paths': "default.log",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'pull_log',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.edr",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'pull_edr',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.trr",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'pull_trr',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.xtc",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'pull_xtc',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.gro",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'pull_gro',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default_pullf.xvg",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'pullf_xvg',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default_pullx.xvg",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'pullx_xvg',
**d }
} ) ]
fw_push = Firework(fts_push,
name = ', '.join(('push', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'push',
**d
}
},
parents = fws_root )
fw_list.append(fw_push)
return fw_list, fw_push
Conversion from LAMMPS data format to PDB¶
The vollowing bash / tcl snippet converts a LAMMPS data file to PDB, assigning the desired names as mapped in a yaml file
#!/bin/bash
# echo "package require jlhvmd; jlh lmp2pdb indenter.lammps indenter.pdb" | vmd -eofexit
vmd -eofexit << 'EOF'
package require jlhvmd
topo readlammpsdata indenter.lammps
jlh type2name SDS_type2name.yaml
jlh name2res SDS_name2res.yaml
set sel [atomselect top all]
$sel writepdb indenter.pdb
EOF
pdb_chain.py indenter.pdb > indenter_wo_chainid.pdb
pdb_reres_by_atom_9999.py indenter_wo_chainid.pdb > indenter_reres.pdb
Requires
- VMD (tested with 1.9.3) with topotools
- jlhvmd VMD plugin: https://github.com/jotelha/jlhvmd
- pdb-tools: https://github.com/haddocking/pdb-tools/
Inspect AFM tip model¶
Read pdb¶
In [17]:
ls $prefix
In [18]:
infile = os.path.join(prefix,'dat','indenter','AU_111_r_25.pdb')
In [19]:
atoms = ase.io.read(infile,format='proteindatabank')
In [20]:
atoms
Out[20]:
Display with ASE view¶
In [21]:
v = view(atoms,viewer='ngl')
v.view._remote_call("setSize", target="Widget", args=["400px", "400px"])
v.view.center()
v.view.background='#ffc'
v
Get the bounding sphere around point set¶
In [25]:
S = atoms.get_positions()
C, R_sq = miniball.get_bounding_ball(S)
R = np.sqrt(R_sq)
del S
xmin = atoms.get_positions().min(axis=0)
xmax = atoms.get_positions().max(axis=0)
In [26]:
C # sphere center
Out[26]:
In [27]:
R # sphere radius
Out[27]:
In [28]:
xmin
Out[28]:
Derive surfactant numbers from sphere dimensions¶
In [29]:
A_Ang = 4*np.pi*R**2 # area in Ansgtrom
A_nm = A_Ang / 10**2
n_per_nm_sq = np.array([0,0.5,1.0,1.5,2.0,2.5,3.0,3.5,4.0,4.5,5.0,5.5,6.0]) # molecules per square nm
N = np.round(A_nm*n_per_nm_sq).astype(int)
In [30]:
A_nm
Out[30]:
In [31]:
N # molecule numbers corresponding to surface concentrations
Out[31]:
Plot 2d projections of point set and bounding sphere¶
In [32]:
# plot side views with sphere projections
plot_side_views_with_spheres(atoms,C,R)
Out[32]:
Plot 3d point set and bounding sphere¶
In [44]:
# bounding sphere surface
u = np.linspace(0, 2 * np.pi, 100)
v = np.linspace(0, np.pi, 100)
us = np.array([
np.outer(np.cos(u), np.sin(v)),
np.outer(np.sin(u), np.sin(v)),
np.outer(np.ones(np.size(u)), np.cos(v))])
bs = C + R*us.T
In [45]:
fig = plt.figure(figsize=(10,10))
ax = fig.add_subplot(111, projection='3d')
x = atoms.get_positions()
ax.scatter(x[:,0], x[:,1],x[:,2], c='y', marker='o')
ax.plot_surface(*bs.T, color='b',alpha=0.1) # bounding sphere
ax.set_xlabel('X')
ax.set_ylabel('Y')
ax.set_zlabel('Z')
plt.show()
Measure surfactant molecule¶
In [46]:
tol = 2 # Ang
Read single surfactant molecule PDB with ParmEd¶
Utilize parmed to read pbd files ASE has difficulties to decipher.
In [168]:
infile = os.path.join(prefix,'1_SDS.pdb')
In [169]:
surfactant_pmd = pmd.load_file(infile)
In [170]:
surfactant_pmd.atoms[-1].atomic_number
Out[170]:
Convert ParmEd structure to ASE atoms¶
In [171]:
surfactant_ase = ase.Atoms(
numbers=[1 if a.atomic_number == 0 else a.atomic_number for a in surfactant_pmd.atoms],
positions=surfactant_pmd.get_coordinates(0))
Get bounding sphere of single surfactant molecule¶
In [172]:
C_surfactant, R_sq_surfactant = miniball.get_bounding_ball(surfactant_ase.get_positions())
In [173]:
C_surfactant
Out[173]:
In [174]:
R_surfactant = np.sqrt(R_sq_surfactant)
In [175]:
R_surfactant
Out[175]:
In [176]:
C_surfactant
Out[176]:
In [177]:
surfactant_ase[:5][1]
Out[177]:
Estimate constraint sphere radii¶
In [178]:
R_OSL = np.linalg.norm(C_surfactant - surfactant_ase[1].position)
In [179]:
R_OSL
Out[179]:
In [180]:
d_head = R_surfactant - R_OSL # roughly: diameter of head group
In [181]:
R_inner = R + tol # place surfactant molecules outside of this sphere
In [182]:
R_inner_constraint = R + tol + d_head # place surfactant tail hydrocarbon within this sphere
In [183]:
R_outer_constraint = R + 2*R_surfactant + tol # place head group sulfur outside this sphere
In [184]:
R_outer = R + 2*R_surfactant + 2*tol # place suractant molecules within this sphere
In [185]:
rr = [R,R_inner,R_inner_constraint,R_outer_constraint,R_outer]
In [186]:
cc = [C]*5
Show 2d projections of geometrical constraints around AFM tip model¶
In [187]:
plot_side_views_with_spheres(atoms,cc,rr,figsize=(20,8))
plt.show()
Packing the surfactant film¶
In [188]:
infile_prefix = os.path.join(prefix,'packmol_infiles')
Identify placeholders in jinja2 template¶
The template looks like this:
In [189]:
with open(os.path.join(infile_prefix,'surfactants_on_sphere.inp'),'r') as f:
print(f.read())
In [190]:
# get all placholders in template
template_file = os.path.join(infile_prefix,'surfactants_on_sphere.inp')
In [191]:
v = find_undeclared_variables(template_file)
In [192]:
v # we want to fill in these placeholder variables
Out[192]:
System and constraint parameters¶
In [193]:
surfactant = 'SDS'
counterion = 'NA'
tolerance = 2 # Ang
sfN = 200
In [194]:
l_surfactant = 2*R_surfactant
In [195]:
# head atom to be geometrically constrained
surfactant_head_bool_ndx = np.array([ a.name == 'S' for a in surfactant_pmd.atoms ],dtype=bool)
In [196]:
# tail atom to be geometrically constrained
surfactant_tail_bool_ndx = np.array([ a.name == 'C12' for a in surfactant_pmd.atoms ],dtype=bool)
In [197]:
head_atom_number = surfactant_head_ndx = np.argwhere(surfactant_head_bool_ndx)[0,0]
In [198]:
tail_atom_number = surfactant_tail_ndx = np.argwhere(surfactant_tail_bool_ndx)[0,0]
In [199]:
# settings can be overridden
packmol_script_context = {
'header': '20191113 TEST PACKING',
'system_name': '200_SDS_on_50_Ang_AFM_tip_model',
'tolerance': tolerance,
'write_restart': True,
'static_components': [
{
'name': 'indenter_reres'
}
]
}
# use pack_sphere function at the notebook's head to generate template context
packmol_script_context.update(
pack_sphere(
C,R_inner_constraint,R_outer_constraint, sfN,
tail_atom_number+1, head_atom_number+1, surfactant, counterion, tolerance))
In [200]:
packmol_script_context # context generated from system and constraint settings
Out[200]:
Fill a packmol input script template with jinja2¶
In [201]:
env = jinja2.Environment()
In [202]:
template = jinja2.Template(open(template_file).read())
In [203]:
rendered = template.render(**packmol_script_context)
In [204]:
rendered_file = os.path.join(prefix,'rendered.inp')
In [205]:
with open(rendered_file,'w') as f:
f.write(rendered)
That's the rendered packmol input file:
In [206]:
print(rendered)
Fail running packmol once¶
In [207]:
packmol = subprocess.Popen(['packmol'],
stdin=subprocess.PIPE,stdout=subprocess.PIPE, stderr=subprocess.PIPE,
cwd=prefix, encoding='utf-8')
In [208]:
outs, errs = packmol.communicate(input=rendered)
In [209]:
print(errs) # error with input from PIPE
Read packmol input from file to avoid obscure Fortran error¶
In [210]:
packmol = subprocess.Popen(['packmol'],
stdin=open(rendered_file,'r'),stdout=subprocess.PIPE, stderr=subprocess.PIPE,
cwd=prefix, encoding='utf-8')
In [211]:
outs, errs = packmol.communicate(input=rendered)
In [212]:
print(outs)
In [213]:
with open('packmol.log','w') as f:
f.write(outs)
Inspect packed systems¶
In [214]:
packmol_pdb = '200_SDS_on_50_Ang_AFM_tip_model_packmol.pdb'
In [215]:
infile = os.path.join(prefix, packmol_pdb)
In [216]:
surfactant_shell_pmd = pmd.load_file(infile)
In [217]:
# with ParmEd and nglview we get automatic bond guessing
pmd_view = nglview.show_parmed(surfactant_shell_pmd)
pmd_view.clear_representations()
pmd_view.background = 'white'
pmd_view.add_representation('ball+stick')
pmd_view
In [218]:
surfactant_shell_ase = ase.Atoms(
numbers=[1 if a.atomic_number == 0 else a.atomic_number for a in surfactant_shell_pmd.atoms],
positions=surfactant_shell_pmd.get_coordinates(0))
In [219]:
# with ASE, we get no bonds at all
ase_view = nglview.show_ase(surfactant_shell_ase)
ase_view.clear_representations()
ase_view.background = 'white'
ase_view.add_representation('ball+stick')
ase_view
Get bounding sphere again and display AFM tip bounding spphere as well as surfactant layer bounding sphere
In [220]:
C_shell, R_sq_shell = miniball.get_bounding_ball(surfactant_shell_ase.get_positions())
In [221]:
C_shell
Out[221]:
In [222]:
R_shell = np.sqrt(R_sq_shell)
In [223]:
R_shell
Out[223]:
In [224]:
plot_side_views_with_spheres(surfactant_shell_ase,[C,C_shell],[R,R_shell])
Out[224]:
In [342]:
surfactant_shell_pmd
Out[342]:
Batch processing: Parametric jobs¶
Provide PACKMOL template¶
In [54]:
project_id = 'juwels-packmol-2020-03-09'
In [17]:
# queries to the data base are simple dictionaries
query = {
'metadata.project': project_id,
}
In [18]:
# use underlying MongoDB functionality to check total number of documents matching query
fp.filepad.count_documents(query)
Out[18]:
In [160]:
infiles = sorted(glob.glob(os.path.join(infile_prefix,'*.inp')))
files = { os.path.basename(f): f for f in infiles }
# metadata common to all these files
metadata = {
'project': project_id,
'type': 'template'
}
fp_files = []
# insert these input files into data base
for name, file_path in files.items():
identifier = '/'.join((project_id,name)) # identifier is like a path on a file system
metadata["name"] = name
fp_files.append( fp.add_file(file_path,identifier=identifier,metadata = metadata) )
In [224]:
# queries to the data base are simple dictionaries
query = {
'metadata.project': project_id,
'metadata.type': 'template'
}
# use underlying MongoDB functionality to check total number of documents matching query
fp.filepad.count_documents(query)
Out[224]:
In [171]:
print(identifier)
In [172]:
# on a lower level, each object has a unique "GridFS id":
pprint(fp_files) # underlying GridFS id and readable identifiers
Provide data files¶
In [272]:
data_prefix = os.path.join(prefix,'packmol_datafiles')
In [273]:
datafiles = sorted(glob.glob(os.path.join(data_prefix,'*')))
files = { os.path.basename(f): f for f in datafiles }
# metadata common to all these files
metadata = {
'project': project_id,
'type': 'data'
}
fp_files = []
# insert these input files into data base
for name, file_path in files.items():
identifier = '/'.join((project_id,name)) # identifier is like a path on a file system
metadata["name"] = name
fp_files.append( fp.add_file(file_path,identifier=identifier,metadata = metadata) )
In [274]:
fp_files
Out[274]:
Span parameter sets¶
In [485]:
machine = 'juwels_devel'
In [480]:
parametric_dimension_labels = ['nmolecules']
In [481]:
parametric_dimensions = [ {
'nmolecules': N } ]
In [458]:
# for testing
parametric_dimensions = [ {
'nmolecules': [N[0]] } ]
In [482]:
parameter_sets = list(
itertools.chain(*[
itertools.product(*list(
p.values())) for p in parametric_dimensions ]) )
parameter_dict_sets = [ dict(zip(parametric_dimension_labels,s)) for s in parameter_sets ]
In [486]:
wf_name = 'PACKMOL {machine:}, {id:}'.format(machine=machine,id=project_id)
fw_name_template = 'nmolecules: {nmolecules:d}'
fw_list = []
fts = [
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': 'surfactants_on_sphere.inp'
},
limit = 1,
new_file_names = ['input.template'] ),
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': 'indenter_reres.pdb'
},
limit = 1,
new_file_names = ['indenter.pdb'] ),
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': '1_SDS.pdb'
},
limit = 1,
new_file_names = ['1_SDS.pdb'] ),
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': '1_NA.pdb'
},
limit = 1,
new_file_names = ['1_NA.pdb'] )
]
files_out = {
'input_file': 'input.template',
'indenter_file': 'indenter.pdb',
'surfatcant_file': '1_SDS.pdb',
'counterion_file': '1_NA.pdb'}
fw_root = Firework(fts,
name = wf_name,
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'input_file_query'
}
}
)
fw_list.append(fw_root)
## Parametric sweep
for d in parameter_dict_sets:
### Template
files_in = {'input_file': 'input.template' }
files_out = { 'input_file': 'input.inp' }
# exports = std_exports[machine].copy()
# Jinja2 context:
packmol_script_context = {
'header': '{:s} packing SDS around AFM probe model'.format(project_id),
'system_name': '{:d}_SDS_on_50_Ang_AFM_tip_model'.format(n),
'tolerance': tolerance,
'write_restart': True,
'static_components': [
{
'name': 'indenter'
}
]
}
# use pack_sphere function at the notebook's head to generate template context
packmol_script_context.update(
pack_sphere(
C,R_inner_constraint,R_outer_constraint, d["nmolecules"],
tail_atom_number+1, head_atom_number+1, surfactant, counterion, tolerance))
ft_template = TemplateWriterTask( {
'context': packmol_script_context,
'template_file': 'input.template',
'template_dir': '.',
'output_file': 'input.inp'} )
fw_template = Firework([ft_template],
name = ', '.join(('template', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'fill_template',
**d
}
},
parents = [ fw_root ] )
fw_list.append(fw_template)
### PACKMOL
files_in = {
'input_file': 'input.inp',
'indenter_file': 'indenter.pdb',
'surfatcant_file': '1_SDS.pdb',
'counterion_file': '1_NA.pdb' }
files_out = {
'data_file': '*_packmol.pdb'}
ft_pack = CmdTask(
cmd='packmol',
opt=['< input.inp'],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True)
fw_pack = Firework([ft_pack],
name = ', '.join(('packmol', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_queue_category'],
'_queueadapter': {
'queue': hpc_max_specs[machine]['queue'],
'walltime' : hpc_max_specs[machine]['walltime'],
'ntasks': 1,
},
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'packmol',
**d
}
},
parents = [ fw_root, fw_template] )
fw_list.append(fw_pack)
### Store
files_in = {'data_file': 'packed.pdb' }
ft_transfer = AddFilesTask( {
'compress': True ,
'paths': "packed.pdb",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'initial_config',
**d }
} )
fw_transfer = Firework([ft_transfer],
name = ', '.join(('transfer', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'transfer',
**d
}
},
parents = [ fw_pack ] )
fw_list.append(fw_transfer)
wf = Workflow(fw_list,
name = wf_name,
metadata = {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'packing'
})
In [488]:
wf.to_file('packing.json')
In [489]:
lp.add_wf(wf)
Out[489]:
Inspect sweep results¶
In [31]:
query = {
"metadata.project": project_id,
}
fp.filepad.count_documents(query)
Out[31]:
In [32]:
query = {
"metadata.project": project_id,
"metadata.type": 'initial_config',
}
fp.filepad.count_documents(query)
Out[32]:
In [33]:
parameter_names = ['nmolecules']
In [34]:
surfactant_shell_pmd_list = []
match_aggregation = {
"$match": query
}
sort_aggregation = {
"$sort": {
"metadata.datetime": pymongo.DESCENDING,
}
}
group_aggregation = {
"$group": {
"_id": { p: '$metadata.{}'.format(p) for p in parameter_names },
"degeneracy": {"$sum": 1}, # number matching data sets
"latest": {"$first": "$gfs_id"} # unique gridfs id of file
}
}
aggregation_pipeline = [ match_aggregation, sort_aggregation, group_aggregation ]
cursor = fp.filepad.aggregate(aggregation_pipeline)
for i, c in enumerate(cursor):
content, metadata = fp.get_file_by_id(c["latest"])
with tempfile.NamedTemporaryFile() as tmp:
tmp.write(content)
surfactant_shell_pmd_list.append(pmd.load_file(tmp.name))
print('.',end='')
print('')
In [80]:
system_selection = surfactant_shell_pmd_list
nx = 3
ny = len(system_selection)
view_labels = ['xy','xz','yz']
molecule_counter = lambda s: np.count_nonzero([r.name == 'SDS' for r in s.residues])
label_generator = lambda i,j: '{:d} SDS, {:s} projection'.format(
molecule_counter(system_selection[i]),view_labels[j])
figsize = (4*nx,4*ny)
fig, axes = plt.subplots(ny,3,figsize=figsize)
for i,system in enumerate(system_selection):
system_ase = ase.Atoms(
numbers=[1 if a.atomic_number == 0 else a.atomic_number for a in system.atoms],
positions=system.get_coordinates(0))
C_shell, R_sq_shell = miniball.get_bounding_ball(system_ase.get_positions())
R_shell = np.sqrt(R_sq_shell)
plot_side_views_with_spheres(
system_ase,[C,C_shell],[R,R_shell],fig=fig,ax=axes[i,:])
for j, ax in enumerate(axes[i,:]):
ax.set_title(label_generator(i,j))
del system_ase
gc.collect()
In [108]:
del fig
del axes
Prepare a Gromacs-processible system¶
In [740]:
gromacs.config.logfilename
Out[740]:
In [741]:
gromacs.environment.flags
Out[741]:
In [742]:
# if true, then stdout and stderr are returned as strings by gromacs wrapper commands
gromacs.environment.flags['capture_output'] = False
In [84]:
print(gromacs.release())
In [85]:
prefix
Out[85]:
In [539]:
system = '200_SDS_on_50_Ang_AFM_tip_model'
pdb = system + '.pdb'
gro = system + '.gro'
top = system + '.top'
posre = system + '.posre.itp'
Tidy up packmol's non-standard pdb¶
In [540]:
# Remove any chain ID from pdb and tidy up
pdb_chain = subprocess.Popen(['pdb_chain',],
stdin=open(packmol_pdb,'r'),stdout=subprocess.PIPE, stderr=subprocess.PIPE,
cwd=prefix, encoding='utf-8')
pdb_tidy = subprocess.Popen(['pdb_tidy',],
stdin=pdb_chain.stdout,stdout=open(pdb,'w'), stderr=subprocess.PIPE,
cwd=prefix, encoding='utf-8')
Generate Gromacs .gro and .top¶
In [541]:
rc,out,err=gromacs.pdb2gmx(
f=pdb,o=gro,p=top,i=posre,ff='charmm36',water='tip3p',
stdout=False,stderr=False)
In [542]:
print(out)
Set simulation box size around system¶
In [543]:
gro_boxed = system + '_boxed.gro'
In [544]:
rc,out,err=gromacs.editconf(
f=gro,o=gro_boxed,d=2.0,bt='cubic',
stdout=False,stderr=False)
In [545]:
print(out)
Batch processing¶
Buld workflow¶
In [43]:
machine = 'juwels_devel'
In [44]:
parametric_dimension_labels = ['nmolecules']
In [45]:
parametric_dimensions = [ {
'nmolecules': N } ]
In [263]:
# for testing
parametric_dimensions = [ {
'nmolecules': [N[0]] } ]
In [264]:
parameter_sets = list(
itertools.chain(*[
itertools.product(*list(
p.values())) for p in parametric_dimensions ]) )
parameter_dict_sets = [ dict(zip(parametric_dimension_labels,s)) for s in parameter_sets ]
In [265]:
source_project_id = 'juwels-packmol-2020-03-09'
project_id = 'juwels-gromacs-prep-2020-03-11'
In [266]:
wf_name = 'GROMACS preparations {machine:}, {id:}'.format(machine=machine,id=project_id)
fw_name_template = 'nmolecules: {nmolecules:d}'
fw_list = []
fts = [
CmdTask(
cmd='echo',
opt=['"Dummy root"'],
store_stdout = False,
store_stderr = False,
use_shell = True,
fizzle_bad_rc= True) ]
files_out = []
fw_root = Firework(fts,
name = wf_name,
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'dummy_root'
}
}
)
fw_list.append(fw_root)
## Parametric sweep
for d in parameter_dict_sets:
### File retrieval
#files_in = {'input_file': 'input.template' }
files_in = {}
files_out = { 'data_file': 'in.pdb' }
# exports = std_exports[machine].copy()
fts_fetch = [ GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'initial_config',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['in.pdb'] ) ]
fw_fetch = Firework(fts_fetch,
name = ', '.join(('fetch', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'fetch',
**d
}
},
parents = [ fw_root ] )
fw_list.append(fw_fetch)
### PDB chain
files_in = {'data_file': 'in.pdb' }
files_out = {'data_file': 'out.pdb'}
fts_pdb_chain = CmdTask(
cmd='pdb_chain',
opt=['< in.pdb > out.pdb'],
store_stdout = False,
store_stderr = False,
use_shell = True,
fizzle_bad_rc= True)
fw_pdb_chain = Firework(fts_pdb_chain,
name = ', '.join(('pdb_chain', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'pdb_chain',
**d
}
},
parents = [ fw_fetch ] )
fw_list.append(fw_pdb_chain)
### PDB tidy
files_in = {'data_file': 'in.pdb' }
files_out = {'data_file': 'out.pdb'}
fts_pdb_tidy = CmdTask(
cmd='pdb_tidy',
opt=['< in.pdb > out.pdb'],
store_stdout = False,
store_stderr = False,
use_shell = True,
fizzle_bad_rc= True)
fw_pdb_tidy = Firework(fts_pdb_tidy,
name = ', '.join(('pdb_tidy', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'pdb_tidy',
**d
}
},
parents = [ fw_pdb_chain ] )
fw_list.append(fw_pdb_tidy)
### GMX pdb2gro
files_in = {'data_file': 'in.pdb' }
files_out = {
'coordinate_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp'}
fts_gmx_pdb2gro = [ CmdTask(
cmd='gmx',
opt=['pdb2gmx',
'-f', 'in.pdb',
'-o', 'default.gro',
'-p', 'default.top',
'-i', 'default.posre.itp',
'-ff', 'charmm36',
'-water' , 'tip3p'],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True) ]
fw_gmx_pdb2gro = Firework(fts_gmx_pdb2gro,
name = ', '.join(('gmx_pdb2gro', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_pdb2gro',
**d
}
},
parents = [ fw_pdb_tidy ] )
fw_list.append(fw_gmx_pdb2gro)
### GMX editconf
files_in = {
'coordinate_file': 'in.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp'}
files_out = {
'coordinate_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp'}
fts_gmx_editconf = [ CmdTask(
cmd='gmx',
opt=['editconf',
'-f', 'in.gro',
'-o', 'default.gro',
'-d', 2.0, # distance between content and box boundary in nm
'-bt', 'cubic', # box type
],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True) ]
fw_gmx_editconf = Firework(fts_gmx_editconf,
name = ', '.join(('gmx_editconf', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_editconf',
**d
}
},
parents = [ fw_gmx_pdb2gro ] )
fw_list.append(fw_gmx_editconf)
### Store
#files_in = {'data_file': 'packed.pdb' }
files_in = {
'coordinate_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp' }
fts_push = [
AddFilesTask( {
'compress': True ,
'paths': "default.gro",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'initial_config_gro',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.top",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'initial_config_top',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "default.posre.itp",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'initial_config_posre_itp',
**d }
} ) ]
fw_push = Firework(fts_push,
name = ', '.join(('push', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'push',
**d
}
},
parents = [ fw_gmx_editconf ] )
fw_list.append(fw_push)
wf = Workflow(fw_list,
name = wf_name,
metadata = {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'gmx_prep'
})
In [267]:
wf.as_dict()
Out[267]:
In [268]:
lp.add_wf(wf)
Out[268]:
In [269]:
wf.to_file('wf-{:s}.yaml'.format(project_id))
Inspect sweep results¶
In [93]:
query = {
"metadata.project": project_id,
}
fp.filepad.count_documents(query)
Out[93]:
In [94]:
query = {
"metadata.project": project_id,
"metadata.type": 'initial_config_gro',
}
fp.filepad.count_documents(query)
Out[94]:
In [95]:
parameter_names = ['nmolecules']
In [96]:
gro_list = []
match_aggregation = {
"$match": query
}
sort_aggregation = {
"$sort": {
"metadata.datetime": pymongo.DESCENDING,
}
}
group_aggregation = {
"$group": {
"_id": { p: '$metadata.{}'.format(p) for p in parameter_names },
"degeneracy": {"$sum": 1}, # number matching data sets
"latest": {"$first": "$gfs_id"} # unique gridfs id of file
}
}
aggregation_pipeline = [ match_aggregation, sort_aggregation, group_aggregation ]
cursor = fp.filepad.aggregate(aggregation_pipeline)
for i, c in enumerate(cursor):
content, metadata = fp.get_file_by_id(c["latest"])
with tempfile.NamedTemporaryFile(suffix='.gro') as tmp:
tmp.write(content)
gro_list.append(pmd.load_file(tmp.name))
print('.',end='')
print('')
In [99]:
gro_list
Out[99]:
In [100]:
# with ParmEd and nglview we get automatic bond guessing
pmd_view = nglview.show_parmed(gro_list[0])
pmd_view.clear_representations()
pmd_view.background = 'white'
pmd_view.add_representation('ball+stick')
pmd_view
Energy minimization with restraints¶
Just to be safe, relax the system a little with positional constraints applied to all ions.
Compile system¶
In [546]:
os.getcwd()
Out[546]:
In [547]:
em_mdp = gromacs.fileformats.MDP('em.mdp.template')
# no change
em_mdp.write('em.mdp')
In [548]:
gmx_grompp = gromacs.grompp.Popen(
f='em.mdp',c=gro,r=gro,o='em.tpr',p=top,
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
out = gmx_grompp.stdout.read()
err = gmx_grompp.stderr.read()
In [549]:
print(err)
In [551]:
print(out)
Run energy minimization¶
In [552]:
gmx_mdrun = gromacs.mdrun.Popen(
deffnm='em',v=True,
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
In [553]:
for line in gmx_mdrun.stdout:
print(line.decode(), end='')
In [ ]:
out = gmx_mdrun.stdout.read()
err = gmx_mdrun.stderr.read()
In [ ]:
print(err)
In [ ]:
print(out)
Energy minimization analysis¶
In [319]:
em_file = 'em.edr'
In [320]:
em_df = panedr.edr_to_df(em_file)
In [554]:
em_df.columns
Out[554]:
In [322]:
fig, ax = plt.subplots(3,2,figsize=(10,12))
em_df.plot('Time','Potential',ax=ax[0,0])
em_df.plot('Time','Pressure',ax=ax[0,1])
em_df.plot('Time','Bond',ax=ax[1,0])
em_df.plot('Time','Position Rest.',ax=ax[1,1])
#em_df.plot('Time','COM Pull En.',ax=ax[1,1])
em_df.plot('Time','Coulomb (SR)',ax=ax[2,0])
em_df.plot('Time','Coul. recip.',ax=ax[2,1])
Out[322]:
In [233]:
mda_trr = mda.Universe(gro,'em.trr')
mda_view = nglview.show_mdanalysis(mda_trr)
mda_view.clear_representations()
mda_view.background = 'white'
mda_view.add_representation('ball+stick')
mda_view
Batch processing¶
Build workflow: em¶
In [42]:
machine = 'juwels_devel'
In [52]:
parametric_dimension_labels = ['nmolecules']
In [53]:
parametric_dimensions = [ {
'nmolecules': N } ]
In [273]:
# for testing
parametric_dimensions = [ {
'nmolecules': [N[0]] } ]
In [274]:
parameter_sets = list(
itertools.chain(*[
itertools.product(*list(
p.values())) for p in parametric_dimensions ]) )
parameter_dict_sets = [ dict(zip(parametric_dimension_labels,s)) for s in parameter_sets ]
In [275]:
source_project_id = 'juwels-gromacs-prep-2020-03-11'
project_id = 'juwels-gromacs-em-2020-03-12'
infile_prefix = 'gmx_em_infiles'
In [276]:
# queries to the data base are simple dictionaries
query = {
'metadata.project': project_id,
}
In [277]:
# use underlying MongoDB functionality to check total number of documents matching query
fp.filepad.count_documents(query)
Out[277]:
In [278]:
infiles = sorted(glob.glob(os.path.join(infile_prefix,'*')))
files = { os.path.basename(f): f for f in infiles }
# metadata common to all these files
metadata = {
'project': project_id,
'type': 'input'
}
fp_files = []
# insert these input files into data base
for name, file_path in files.items():
identifier = '/'.join((project_id,name)) # identifier is like a path on a file system
metadata["name"] = name
fp_files.append( fp.add_file(file_path,identifier=identifier,metadata = metadata) )
In [291]:
wf_name = 'GROMACS energy minimization {machine:}, {id:}'.format(machine=machine,id=project_id)
fw_name_template = 'nmolecules: {nmolecules:d}'
fw_list = []
files_out = { 'input_file': 'default.mdp' }
fts_root = [
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': 'em.mdp'
},
limit = 1,
new_file_names = ['default.mdp'] ) ]
fw_root = Firework(fts_root,
name = wf_name,
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'input_file_query'
}
}
)
fw_list.append(fw_root)
## Parametric sweep
for d in parameter_dict_sets:
### File retrieval
#files_in = {'input_file': 'input.template' }
files_in = {}
files_out = {
'coordinate_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp',
}
fts_fetch = [
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'initial_config_gro',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.gro'] ),
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'initial_config_top',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.top'] ),
GetFilesByQueryTask(
query = {
'metadata->project': source_project_id,
'metadata->type': 'initial_config_posre_itp',
'metadata->nmolecules': d["nmolecules"]
},
sort_key = 'metadata.datetime',
sort_direction = pymongo.DESCENDING,
limit = 1,
new_file_names = ['default.posre.itp'] ) ]
fw_fetch = Firework(fts_fetch,
name = ', '.join(('fetch', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'fetch',
**d
}
},
parents = [ fw_root ] )
fw_list.append(fw_fetch)
### GMX grompp
files_in = {
'input_file': 'default.mdp',
'coordinate_file': 'default.gro',
'topology_file': 'default.top',
'restraint_file': 'default.posre.itp'}
files_out = {
'input_file': 'default.tpr',
'parameter_file': 'mdout.mdp' }
fts_gmx_grompp = [ CmdTask(
cmd='gmx',
opt=['grompp',
'-f', 'default.mdp',
'-c', 'default.gro',
'-r', 'default.gro',
'-o', 'default.tpr',
'-p', 'default.top',
],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True) ]
fw_gmx_grompp = Firework(fts_gmx_grompp,
name = ', '.join(('gmx_grompp_em', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_grompp_em',
**d
}
},
parents = [ fw_fetch, fw_root ] )
fw_list.append(fw_gmx_grompp)
### GMX mdrun
files_in = {'input_file': 'em.tpr'}
files_out = {
'energy_file': 'em.edr',
'trajectory_file': 'em.trr',
'final_config': 'em.gro' }
fts_gmx_mdrun = [ CmdTask(
cmd='gmx',
opt=[' mdrun',
'-deffnm', 'em', '-v' ],
stderr_file = 'std.err',
stdout_file = 'std.out',
store_stdout = True,
store_stderr = True,
use_shell = True,
fizzle_bad_rc= True) ]
fw_gmx_mdrun = Firework(fts_gmx_mdrun,
name = ', '.join(('gmx_mdrun_em', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_queue_category'],
'_queueadapter': {
'queue': hpc_max_specs[machine]['queue'],
'walltime' : hpc_max_specs[machine]['walltime'],
'ntasks': 96,
},
'_files_in': files_in,
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'gmx_mdrun_em',
**d
}
},
parents = [ fw_gmx_grompp ] )
fw_list.append(fw_gmx_mdrun)
# Store results
files_in = {
'energy_file': 'em.edr',
'trajectory_file': 'em.trr',
'final_config': 'em.gro' }
files_out = {}
fts_push = [
AddFilesTask( {
'compress': True ,
'paths': "em.edr",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'em_edr',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "em.trr",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'em_trr',
**d }
} ),
AddFilesTask( {
'compress': True ,
'paths': "em.gro",
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'em_gro',
**d }
} ) ]
fw_push = Firework(fts_push,
name = ', '.join(('push', fw_name_template.format(**d))),
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_in': files_in,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'push',
**d
}
},
parents = [ fw_gmx_mdrun ] )
fw_list.append(fw_push)
wf = Workflow(fw_list,
name = wf_name,
metadata = {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'gmx_prep'
})
In [292]:
wf.to_file('wf-{:s}.yaml'.format(project_id))
In [294]:
lp.add_wf(wf)
Out[294]:
Build workflow: prep & em¶
In [450]:
machine = 'juwels_devel'
In [451]:
parametric_dimension_labels = ['nmolecules']
In [452]:
parametric_dimensions = [ {
'nmolecules': N } ]
In [453]:
# for testing
parametric_dimensions = [ {
'nmolecules': [N[0]] } ]
In [454]:
parameter_sets = list(
itertools.chain(*[
itertools.product(*list(
p.values())) for p in parametric_dimensions ]) )
parameter_dict_sets = [ dict(zip(parametric_dimension_labels,s)) for s in parameter_sets ]
In [411]:
source_project_id = 'juwels-packmol-2020-03-09'
project_id = 'juwels-gromacs-em-2020-03-12'
infile_prefix = 'gmx_em_infiles'
In [412]:
# queries to the data base are simple dictionaries
query = {
'metadata.project': project_id,
}
In [413]:
# use underlying MongoDB functionality to check total number of documents matching query
fp.filepad.count_documents(query)
Out[413]:
In [414]:
infiles = sorted(glob.glob(os.path.join(infile_prefix,'*')))
files = { os.path.basename(f): f for f in infiles }
# metadata common to all these files
metadata = {
'project': project_id,
'type': 'input'
}
fp_files = []
# insert these input files into data base
for name, file_path in files.items():
identifier = '/'.join((project_id,name)) # identifier is like a path on a file system
metadata["name"] = name
fp_files.append( fp.add_file(file_path,identifier=identifier,metadata = metadata) )
In [415]:
wf_name = 'GROMACS preparations & energy minimization, {machine:}, {id:}'.format(machine=machine,id=project_id)
fw_name_template = 'nmolecules: {nmolecules:d}'
fw_list = []
fts_root_dummy = [
CmdTask(
cmd='echo',
opt=['"Dummy root"'],
store_stdout = False,
store_stderr = False,
use_shell = True,
fizzle_bad_rc= True) ]
files_out = {}
fw_root_dummy = Firework(fts_root_dummy,
name = wf_name,
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'dummy_root'
}
}
)
fw_list.append(fw_root_dummy)
files_out = { 'input_file': 'default.mdp' }
fts_root_pull = [
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': 'em.mdp'
},
limit = 1,
new_file_names = ['default.mdp'] ) ]
fw_root_pull = Firework(fts_root_pull,
name = wf_name,
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'input_file_query'
}
}
)
fw_list.append(fw_root_pull)
## Parametric sweep
for d in parameter_dict_sets:
fw_list_tmp, fw_gmx_prep_pull_leaf = sub_wf_gmx_prep_pull(d, [fw_root_dummy])
fw_list.extend(fw_list_tmp)
fw_list_tmp, fw_gmx_prep_leaf = sub_wf_gmx_prep(d, [fw_gmx_prep_pull_leaf])
fw_list.extend(fw_list_tmp)
fw_list_tmp, _ = sub_wf_gmx_prep_push(d, [fw_gmx_prep_leaf])
fw_list.extend(fw_list_tmp)
fw_list_tmp, fw_gmx_mdrun = sub_wf_gmx_em(d, [fw_root_pull,fw_gmx_prep_leaf])
fw_list.extend(fw_list_tmp)
fw_list_tmp, fw_gmx_mdrun_push = sub_wf_gmx_em_psuh(d, [fw_gmx_mdrun])
fw_list.extend(fw_list_tmp)
wf = Workflow(fw_list,
name = wf_name,
metadata = {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'gmx_prep'
})
In [440]:
fw_list_tmp, fw_gmx_mdrun_push = sub_wf_gmx_em_push(d, None)
#fw_list.extend(fw_list_tmp)
In [441]:
len(fw_list_tmp)
Out[441]:
In [442]:
sub_wf = Workflow(fw_list_tmp)
In [449]:
lp.append_wf(sub_wf,[24198])
In [447]:
wf.as_dict()
Out[447]:
In [417]:
wf.to_file('wf-{:s}.yaml'.format(project_id))
In [418]:
lp.add_wf(wf)
Out[418]:
Inspect sweep results¶
In [461]:
query = {
"metadata.project": project_id,
"metadata.type": 'em_edr',
}
fp.filepad.count_documents(query)
Out[461]:
In [462]:
parameter_names = ['nmolecules']
In [532]:
em_list = []
match_aggregation = {
"$match": query
}
sort_aggregation = {
"$sort": {
"metadata.datetime": pymongo.DESCENDING,
}
}
group_aggregation = {
"$group": {
"_id": { p: '$metadata.{}'.format(p) for p in parameter_names },
"degeneracy": {"$sum": 1}, # number matching data sets
"latest": {"$first": "$gfs_id"} # unique gridfs id of file
}
}
aggregation_pipeline = [ match_aggregation, sort_aggregation, group_aggregation ]
cursor = fp.filepad.aggregate(aggregation_pipeline)
for i, c in enumerate(cursor):
content, metadata = fp.get_file_by_id(c["latest"])
with tempfile.NamedTemporaryFile(suffix='.edr') as tmp:
tmp.write(content)
em_df = panedr.edr_to_df(tmp.name)
mi = pd.MultiIndex.from_product(
[[int(metadata["metadata"]["nmolecules"])],em_df.index],
names=['nmolecules','step'])
em_mi_df = em_df.set_index(mi)
em_list.append(em_mi_df)
print('.',end='')
print('')
em_df = pd.concat(em_list)
In [537]:
fig, ax = plt.subplots(3,2,figsize=(10,12))
em_df.plot('Time','Potential',ax=ax[0,0])
em_df.plot('Time','Pressure',ax=ax[0,1])
em_df.plot('Time','Bond',ax=ax[1,0])
#em_df.plot('Time','Position Rest.',ax=ax[1,1])
#em_df.plot('Time','COM Pull En.',ax=ax[1,1])
em_df.plot('Time','Coulomb (SR)',ax=ax[2,0])
em_df.plot('Time','Coul. recip.',ax=ax[2,1])
Out[537]:
In [559]:
query = {
"metadata.project": project_id,
"metadata.type": { '$in': ['em_trr','em_gro'] },
}
fp.filepad.count_documents(query)
Out[559]:
In [731]:
# Building a rather sophisticated aggregation pipeline
parameter_names = ['nmolecules', 'type']
query = {
"metadata.project": project_id,
"metadata.type": { '$in': ['em_trr','em_gro'] },
}
aggregation_pipeline = []
aggregation_pipeline.append({
"$match": query
})
aggregation_pipeline.append({
"$sort": {
"metadata.nmolecules": pymongo.ASCENDING,
"metadata.datetime": pymongo.DESCENDING,
}
})
aggregation_pipeline.append({
"$group": {
"_id": { p: '$metadata.{}'.format(p) for p in parameter_names },
"degeneracy": {"$sum": 1}, # number matching data sets
"latest": {"$first": "$gfs_id"} # unique gridfs id of file
}
})
parameter_names = ['nmolecules']
aggregation_pipeline.append({
"$group": {
"_id": { p: '$_id.{}'.format(p) for p in parameter_names },
"type": {"$addToSet": "$_id.type"},
"gfs_id": {"$addToSet": "$latest"}
#"$_id.type": "$latest"
}
})
aggregation_pipeline.append({
'$project': {
'_id': False,
**{ p: '$_id.{}'.format(p) for p in parameter_names},
'objects': {
'$zip': {
'inputs': [ '$type', '$gfs_id' ],
'useLongestLength': True,
'defaults': [None,None]
}
}
}
})
aggregation_pipeline.append({
'$project': {
**{ p: True for p in parameter_names},
'objects': {'$arrayToObject': '$objects'}
#'objects': False
}
})
aggregation_pipeline.append({
'$addFields': {
'objects': { **{ p: '${}'.format(p) for p in parameter_names} }
}
})
aggregation_pipeline.append({
'$replaceRoot': { 'newRoot': '$objects' }
})
# display results with
# for i, c in enumerate(cursor):
# print(c)
# yields documents in the form
# {'em_gro': '5e6a4e3d6c26f976ceae5e38', 'em_trr': '5e6a4e3a6c26f976ceae5e14', 'nmolecules': '44'}
# i.e. most recent topology file and trajectory file per concentration
cursor = fp.filepad.aggregate(aggregation_pipeline)
In [732]:
mda_trr_list = []
for i, c in enumerate(cursor):
em_gro_content, _ = fp.get_file_by_id(c["em_gro"])
em_trr_content, _ = fp.get_file_by_id(c["em_trr"])
# STream approach won't work
# with io.TextIOWrapper( io.BytesIO(em_gro_content) ) as gro, \
# io.BytesIO(em_trr_content) as trr:
#mda_trr_list.append(
# mda.Universe(
# gro,trr, topology_format = 'GRO', format='TRR') )
with tempfile.NamedTemporaryFile(suffix='.gro') as gro, \
tempfile.NamedTemporaryFile(suffix='.trr') as trr:
gro.write(em_gro_content)
trr.write(em_trr_content)
mda_trr_list.append( mda.Universe(gro.name,trr.name) )
print('.',end='')
print('')
In [733]:
mda_trr = mda_trr_list[0]
mda_view = nglview.show_mdanalysis(mda_trr)
mda_view.clear_representations()
mda_view.background = 'white'
mda_view.add_representation('ball+stick')
mda_view
Batch processing: packing, prep & em¶
In [785]:
machine = 'juwels_devel'
In [786]:
parametric_dimension_labels = ['nmolecules']
In [787]:
N
Out[787]:
In [788]:
parametric_dimensions = [ {
'nmolecules': N } ]
In [789]:
# for testing
parametric_dimensions = [ {
'nmolecules': [N[-1]] } ]
In [790]:
parameter_sets = list(
itertools.chain(*[
itertools.product(*list(
p.values())) for p in parametric_dimensions ]) )
parameter_dict_sets = [ dict(zip(parametric_dimension_labels,s)) for s in parameter_sets ]
In [791]:
# source_project_id = 'juwels-packmol-2020-03-09'
project_id = 'juwels-afm-probe-solvation-trial-a-2020-03-13'
# infile_prefix = 'gmx_em_infiles'
In [792]:
# queries to the data base are simple dictionaries
query = {
'metadata.project': project_id,
}
In [793]:
# use underlying MongoDB functionality to check total number of documents matching query
fp.filepad.count_documents(query)
Out[793]:
Provide PACKMOL template¶
In [794]:
infile_prefix = os.path.join(prefix,'packmol_infiles')
infiles = sorted(glob.glob(os.path.join(infile_prefix,'*.inp')))
files = { os.path.basename(f): f for f in infiles }
# metadata common to all these files
metadata = {
'project': project_id,
'type': 'template'
}
fp_files = []
# insert these input files into data base
for name, file_path in files.items():
identifier = '/'.join((project_id,name)) # identifier is like a path on a file system
metadata["name"] = name
fp_files.append( fp.add_file(file_path,identifier=identifier,metadata = metadata) )
In [795]:
# queries to the data base are simple dictionaries
query = {
'metadata.project': project_id,
'metadata.type': 'template'
}
# use underlying MongoDB functionality to check total number of documents matching query
fp.filepad.count_documents(query)
Out[795]:
In [796]:
print(identifier)
In [797]:
# on a lower level, each object has a unique "GridFS id":
pprint(fp_files) # underlying GridFS id and readable identifiers
Provide data files¶
In [798]:
data_prefix = os.path.join(prefix,'packmol_datafiles')
In [799]:
datafiles = sorted(glob.glob(os.path.join(data_prefix,'*')))
files = { os.path.basename(f): f for f in datafiles }
# metadata common to all these files
metadata = {
'project': project_id,
'type': 'data'
}
fp_files = []
# insert these input files into data base
for name, file_path in files.items():
identifier = '/'.join((project_id,name)) # identifier is like a path on a file system
metadata["name"] = name
fp_files.append( fp.add_file(file_path,identifier=identifier,metadata = metadata) )
Provide energy minimization input files¶
In [800]:
infile_prefix = 'gmx_em_infiles'
infiles = sorted(glob.glob(os.path.join(infile_prefix,'*')))
files = { os.path.basename(f): f for f in infiles }
# metadata common to all these files
metadata = {
'project': project_id,
'type': 'input'
}
fp_files = []
# insert these input files into data base
for name, file_path in files.items():
identifier = '/'.join((project_id,name)) # identifier is like a path on a file system
metadata["name"] = name
fp_files.append( fp.add_file(file_path,identifier=identifier,metadata = metadata) )
Build workflow¶
In [809]:
wf_name = 'pack, preparations & energy minimization, {machine:}, {id:}'.format(machine=machine,id=project_id)
fw_name_template = 'nmolecules: {nmolecules:d}'
fw_list = []
# sub-wf pack root
fts_root_pack = [
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': 'surfactants_on_sphere.inp'
},
limit = 1,
new_file_names = ['input.template'] ),
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': 'indenter_reres.pdb'
},
limit = 1,
new_file_names = ['indenter.pdb'] ),
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': '1_SDS.pdb'
},
limit = 1,
new_file_names = ['1_SDS.pdb'] ),
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': '1_NA.pdb'
},
limit = 1,
new_file_names = ['1_NA.pdb'] )
]
files_out = {
'input_file': 'input.template',
'indenter_file': 'indenter.pdb',
'surfatcant_file': '1_SDS.pdb',
'counterion_file': '1_NA.pdb'}
fw_root_pack = Firework(fts_root_pack,
name = wf_name,
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'subwf': 'pack',
'step': 'input_file_query'
}
}
)
fw_list.append(fw_root_pack)
# sub-wf gmx prep root
fts_root_prep = [
CmdTask(
cmd='echo',
opt=['"Dummy root"'],
store_stdout = False,
store_stderr = False,
use_shell = True,
fizzle_bad_rc= True) ]
files_out = {}
fw_root_prep = Firework(fts_root_prep,
name = wf_name,
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'subwf': 'prep',
'step': 'dummy_root'
}
}
)
fw_list.append(fw_root_prep)
# sub-wf gmx em root
files_out = { 'input_file': 'default.mdp' }
fts_root_em = [
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': 'em.mdp'
},
limit = 1,
new_file_names = ['default.mdp'] ) ]
fw_root_em = Firework(fts_root_em,
name = wf_name,
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'subwf': 'em',
'step': 'input_file_query'
}
}
)
fw_list.append(fw_root_em)
## Parametric sweep
for d in parameter_dict_sets:
# pack
fw_list_tmp, fw_pack_leaf = sub_wf_pack(d, [fw_root_pack])
fw_list.extend(fw_list_tmp)
fw_list_tmp, _ = sub_wf_pack_push(d, [fw_pack_leaf])
fw_list.extend(fw_list_tmp)
# prep
fw_list_tmp, fw_prep_leaf = sub_wf_gmx_prep(d, [fw_pack_leaf])
fw_list.extend(fw_list_tmp)
fw_list_tmp, _ = sub_wf_gmx_prep_push(d, [fw_prep_leaf])
fw_list.extend(fw_list_tmp)
# em
fw_list_tmp, fw_gmx_mdrun = sub_wf_gmx_em(d, [fw_root_em,fw_prep_leaf])
fw_list.extend(fw_list_tmp)
fw_list_tmp, fw_gmx_mdrun_push = sub_wf_gmx_em_push(d, [fw_gmx_mdrun])
fw_list.extend(fw_list_tmp)
wf = Workflow(fw_list,
name = wf_name,
metadata = {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'pack, prep, em'
})
In [812]:
wf.to_file('wf-{:s}.json'.format(project_id))
In [813]:
lp.add_wf(wf)
Out[813]:
Pulling¶
Utilize harmonic pulling to attach surfactants to substrate closely.
Create index groups for pulling¶
In [37]:
#pdb = '200_SDS_on_50_Ang_AFM_tip_model.pdb'
gro = 'em.gro'
top = 'sys.top'
ndx = 'standard.ndx'
In [743]:
import parmed as pmd
pmd_top_gro = pmd.gromacs.GromacsTopologyFile(top)
#pmd_top_pdb = pmd.gromacs.GromacsTopologyFile(top)
pmd_gro = pmd.gromacs.GromacsGroFile.parse(gro)
pmd_top_gro.box = pmd_gro.box
pmd_top_gro.positions = pmd_gro.positions
#pmd_pdb = pmd.formats.pdb.PDBFile.parse(pdb)
#pmd_top_pdb.box = pmd_pdb.box
#pmd_top_pdb.positions = pmd_pdb.positions
In [327]:
tail_atom_ndx = np.array([
i+1 for i,a in enumerate(pmd_top_gro.atoms) if a.name == 'C12' and a.residue.name == 'SDS' ])
# gromacs ndx starts at 1
In [814]:
tail_atom_ndx
Generate standard index file for system¶
In [823]:
gmx_make_ndx = gromacs.make_ndx.Popen(
f=gro,o=ndx,
input='q',
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
In [824]:
out_str, err_str = gmx_make_ndx.communicate()
In [825]:
print(out_str)
In [826]:
print(err_str)
Enhance standard index file by pulling groups¶
In [333]:
pull_groups_ndx_in = gromacs.fileformats.NDX(ndx)
In [334]:
pull_groups_ndx_out = gromacs.fileformats.NDX()
In [335]:
for i, a in enumerate(tail_atom_ndx):
pull_group_name = 'pull_group_{:04d}'.format(i)
pull_groups_ndx_out[pull_group_name] = a
In [336]:
pull_groups_ndx_in.update(pull_groups_ndx_out)
In [337]:
pull_groups_ndx_in.write('pull_groups.ndx')
Create mdp input file with pulling groups and coordinates¶
In [445]:
# gromacs wrapper parses mdp files
pull_mdp = gromacs.fileformats.MDP('pull.mdp.template')
pull_mdp['nsteps'] = 10000
N_pull_coords = len(pull_groups_ndx_out)
pull_mdp['pull_ncoords'] = N_pull_coords
pull_mdp['pull_ngroups'] = N_pull_coords + 1
pull_mdp['pull_group1_name'] = 'Substrate' # the reference group
for i, n in enumerate(pull_groups_ndx_out):
pull_mdp["pull_group{:d}_name".format(i+2)] = n
pull_mdp["pull_coord{:d}_type".format(i+1)] = 'umbrella' # harmonic potential
pull_mdp["pull_coord{:d}_geometry".format(i+1)] = 'distance' # simple distance increase
pull_mdp["pull_coord{:d}_dim".format(i+1)] = 'Y Y Y' # pull in all directions
pull_mdp["pull_coord{:d}_groups".format(i+1)] = "1 {:d}".format(i+2) # groups 1 (Chain A) and 2 (Chain B) define the reaction coordinate
pull_mdp["pull_coord{:d}_start".format(i+1)] = 'yes' # define initial COM distance > 0
pull_mdp["pull_coord{:d}_rate".format(i+1)] = -0.1 # 0.1 nm per ps = 10 nm per ns
pull_mdp["pull_coord{:d}_k".format(i+1)] = 1000 # kJ mol^-1 nm^-2
pull_mdp.write('pull.mdp')
Compile system¶
In [446]:
gmx_grompp = gromacs.grompp.Popen(
f='pull.mdp',n='pull_groups.ndx',c=gro,r=gro,o='pull.tpr',p=top,
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
out = gmx_grompp.stdout.read()
err = gmx_grompp.stderr.read()
In [447]:
print(err)
In [448]:
print(out)
Run pulling simulation¶
In [449]:
gmx_mdrun = gromacs.mdrun.Popen(
deffnm='pull',v=True,
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
In [450]:
out = gmx_mdrun.stdout.read()
err = gmx_mdrun.stderr.read()
In [451]:
print(err)
In [779]:
print(out)
Batch processing¶
In [59]:
# cast into parameters
surfactant = 'SDS'
tail_atom_name = 'C12'
substrate = 'AUM'
counterion = 'NA'
nsubstrate = 3873
Build workflow: pull¶
In [190]:
machine = 'juwels_devel'
In [191]:
parametric_dimension_labels = ['nmolecules']
In [192]:
parametric_dimensions = [ {
'nmolecules': N } ]
In [240]:
# for testing
parametric_dimensions = [ {
'nmolecules': [N[1]] } ]
In [243]:
parameter_sets = list(
itertools.chain(*[
itertools.product(*list(
p.values())) for p in parametric_dimensions ]) )
parameter_dict_sets = [ dict(zip(parametric_dimension_labels,s)) for s in parameter_sets ]
In [244]:
source_project_id = 'juwels-gromacs-prep-2020-03-11'
project_id = 'juwels-pull-2020-03-17'
infile_prefix = 'gmx_infiles'
In [245]:
# queries to the data base are simple dictionaries
query = {
'metadata.project': project_id,
}
In [246]:
# use underlying MongoDB functionality to check total number of documents matching query
fp.filepad.count_documents(query)
Out[246]:
In [247]:
infiles = sorted(glob.glob(os.path.join(infile_prefix,'*')))
files = { os.path.basename(f): f for f in infiles }
# metadata common to all these files
metadata = {
'project': project_id,
'type': 'input'
}
fp_files = []
# insert these input files into data base
for name, file_path in files.items():
identifier = '/'.join((project_id,name)) # identifier is like a path on a file system
metadata["name"] = name
fp_files.append( fp.add_file(file_path,identifier=identifier,metadata = metadata) )
In [248]:
fp_files
Out[248]:
In [272]:
wf_name = 'GROMACS surfactant pulling {machine:}, {id:}'.format(machine=machine,id=project_id)
fw_name_template = 'nmolecules: {nmolecules:d}'
fw_list = []
files_out = {
'template_file': 'sys.top.template',
'parameter_file': 'pull.mdp.template',
}
fts_root_pull = [
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': 'sys.top.template',
},
limit = 1,
new_file_names = ['sys.top.template'] ),
GetFilesByQueryTask(
query = {
'metadata->project': project_id,
'metadata->name': 'pull.mdp.template',
},
limit = 1,
new_file_names = ['pull.mdp.template'] ) ]
fw_root_pull = Firework(fts_root_pull,
name = wf_name,
spec = {
'_category': hpc_max_specs[machine]['fw_noqueue_category'],
'_files_out': files_out,
'metadata': {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'step': 'input_file_query'
}
}
)
fw_list.append(fw_root_pull)
## Parametric sweep
for d in parameter_dict_sets:
# pull prep pull files
fw_list_tmp, fw_pull_prep_pull_leaf = sub_wf_pull_prep_pull(d, [])
fw_list.extend(fw_list_tmp)
# pull prep
fw_list_tmp, fw_pull_prep_leaf = sub_wf_pull_prep(d, [fw_root_pull,fw_pull_prep_pull_leaf])
fw_list.extend(fw_list_tmp)
# pull prep push files
fw_list_tmp, _ = sub_wf_pull_prep_push(d, [fw_pull_prep_leaf])
fw_list.extend(fw_list_tmp)
# pull
fw_list_tmp, fw_pull_leaf = sub_wf_gmx_pull(d, [fw_pull_prep_leaf])
fw_list.extend(fw_list_tmp)
# pull push files
fw_list_tmp, _ = sub_wf_gmx_pull_push(d, [fw_pull_leaf])
fw_list.extend(fw_list_tmp)
wf = Workflow(fw_list,
name = wf_name,
metadata = {
'project': project_id,
'datetime': str(datetime.datetime.now()),
'type': 'pull'
})
In [273]:
wf.to_file('wf-{:s}.yaml'.format(project_id))
In [274]:
lp.add_wf(wf)
Out[274]:
Pulling analysis¶
Global observables¶
The gmx energy table:
1 Restraint-Pot. 2 U-B 3 Proper-Dih. 4 LJ-14
5 Coulomb-14 6 LJ-(SR) 7 Coulomb-(SR) 8 Coul.-recip.
9 Position-Rest. 10 COM-Pull-En. 11 Potential 12 Kinetic-En.
13 Total-Energy 14 Temperature 15 Pressure 16 Constr.-rmsd
17 Vir-XX 18 Vir-XY 19 Vir-XZ 20 Vir-YX
21 Vir-YY 22 Vir-YZ 23 Vir-ZX 24 Vir-ZY
25 Vir-ZZ 26 Pres-XX 27 Pres-XY 28 Pres-XZ
29 Pres-YX 30 Pres-YY 31 Pres-YZ 32 Pres-ZX
33 Pres-ZY 34 Pres-ZZ 35 #Surf*SurfTen 36 T-restconverted to dict with regex
\s+([0-9]+)\s+([^\s]+)
and replacement
'$2': $1,\nIn [420]:
gmx_energy_dict = {
'Restraint-Pot.': 1,
'U-B': 2,
'Proper-Dih.': 3,
'LJ-14': 4,
'Coulomb-14': 5,
'LJ-(SR)': 6,
'Coulomb-(SR)': 7,
'Coul.-recip.': 8,
'Position-Rest.': 9,
'COM-Pull-En.': 10,
'Potential': 11,
'Kinetic-En.': 12,
'Total-Energy': 13,
'Temperature': 14,
'Pressure': 15,
'Constr.-rmsd': 16,
'Vir-XX': 17,
'Vir-XY': 18,
'Vir-XZ': 19,
'Vir-YX': 20,
'Vir-YY': 21,
'Vir-YZ': 22,
'Vir-ZX': 23,
'Vir-ZY': 24,
'Vir-ZZ': 25,
'Pres-XX': 26,
'Pres-XY': 27,
'Pres-XZ': 28,
'Pres-YX': 29,
'Pres-YY': 30,
'Pres-YZ': 31,
'Pres-ZX': 32,
'Pres-ZY': 33,
'Pres-ZZ': 34,
'#Surf*SurfTen': 35,
'T-rest': 36,
}
In [429]:
query = {
"metadata.project": project_id,
"metadata.type": 'pull_edr',
#"metadata.nmolecules": '44'
}
fp.filepad.count_documents(query)
Out[429]:
In [430]:
parameter_names = ['nmolecules']
In [431]:
gmx_energy_selection = [
'Restraint-Pot.',
'Position-Rest.',
'COM-Pull-En.',
'Potential',
'Kinetic-En.',
'Total-Energy',
'Temperature',
'Pressure',
'Constr.-rmsd',
]
In [508]:
res_list = []
failed_list = []
match_aggregation = {
"$match": query
}
sort_aggregation = {
"$sort": {
"metadata.datetime": pymongo.DESCENDING,
}
}
group_aggregation = {
"$group": {
"_id": { p: '$metadata.{}'.format(p) for p in parameter_names },
"degeneracy": {"$sum": 1}, # number matching data sets
"latest": {"$first": "$gfs_id"} # unique gridfs id of file
}
}
aggregation_pipeline = [ match_aggregation, sort_aggregation, group_aggregation ]
cursor = fp.filepad.aggregate(aggregation_pipeline)
for i, c in enumerate(cursor):
content, metadata = fp.get_file_by_id(c["latest"])
nmolecules = int(metadata["metadata"]["nmolecules"])
#df = panedr.edr_to_df(tmp.name), fails
tmpin = tempfile.NamedTemporaryFile(mode='w+b',suffix='.edr', delete=False)
# cur_res_dict = {}
with tmpin:
tmpin.write(content)
#tmpin.seek(0)
for sel in gmx_energy_selection:
try:
tmpout = tempfile.NamedTemporaryFile(suffix='.xvg', delete=False)
res = gromacs.energy(f=tmpin.name,o=tmpout.name,
input=str(gmx_energy_dict[sel]))
#with open(tmpout.name,'r') as f:
# xvg = f.read()
#tmpout.delete()
xvg = mda.auxiliary.XVG.XVGReader(tmpout.name)
xvg_time = xvg.read_all_times()
xvg_data = np.array([ f.data[1:] for f in xvg ]).flatten() # 1st entry contains times
os.unlink(tmpout.name)
except:
logger.warning("Failed to read '{:s}' from data set {:d}.".format(sel,i))
failed_list.append((nmolecules, sel))
else:
res_list.append({'nmolecules': [nmolecules]*len(xvg_time), 'time': xvg_time, sel: xvg_data})
#[[int(metadata["metadata"]["nmolecules"])],df.index]
#mi = pd.MultiIndex.from_product(
# [[int(metadata["metadata"]["nmolecules"])],df.index],
# names=['nmolecules','step'])
#mi_df = df.set_index(mi)
#res_list.append(mi_df)
#tmpin.delete()
os.unlink(tmpin.name)
print('.',end='')
print('')
#res_df = pd.concat(res_list)
In [538]:
res_df = None
for r in res_list:
#nmolecules = r['nmolecules']
#data={key: value for key, value in r.items() if key != 'nmolecules'}
# columns=[key if key == 'time' else (nmolecules,key) for key in data]
cur_df = pd.DataFrame(r)
if res_df is None:
res_df = cur_df
else:
res_df = pd.merge(res_df, cur_df, how='outer', on=['nmolecules', 'time'])
res_df_mi = res_df.set_index(['nmolecules','time'])
In [539]:
res_df
Out[539]:
In [544]:
fig, ax = plt.subplots(3,2,figsize=(10,12))
res_df.plot('time','Potential',ax=ax[0,0])
res_df.plot('time','Pressure',ax=ax[0,1])
res_df.plot('time','Restraint-Pot.',ax=ax[1,0])
res_df.plot('time','Position-Rest.',ax=ax[1,1])
res_df.plot('time','COM-Pull-En.',ax=ax[2,0])
#res_df.plot('Coulomb (SR)',ax=ax[2,0])
#res_df.plot('Coul. recip.',ax=ax[2,1])
Out[544]:
Pulling forces¶
In [567]:
res_list = []
failed_list = []
query = {
"metadata.project": project_id,
"metadata.type": 'pullf_xvg',
}
fp.filepad.count_documents(query)
match_aggregation = {
"$match": query
}
sort_aggregation = {
"$sort": {
"metadata.datetime": pymongo.DESCENDING,
}
}
group_aggregation = {
"$group": {
"_id": { p: '$metadata.{}'.format(p) for p in parameter_names },
"degeneracy": {"$sum": 1}, # number matching data sets
"latest": {"$first": "$gfs_id"} # unique gridfs id of file
}
}
aggregation_pipeline = [ match_aggregation, sort_aggregation, group_aggregation ]
cursor = fp.filepad.aggregate(aggregation_pipeline)
for i, c in enumerate(cursor):
content, metadata = fp.get_file_by_id(c["latest"])
nmolecules = int(metadata["metadata"]data:image/png;base64,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["nmolecules"])
tmpin = tempfile.NamedTemporaryFile(mode='w+b',suffix='.xvg', delete=False)
with tmpin:
tmpin.write(content)
try:
xvg = mda.auxiliary.XVG.XVGReader(tmpin.name)
xvg_time = xvg.read_all_times()
xvg_data = np.array([ f.data[1:] for f in xvg ])# .flatten() # 1st entry contains times
except:
logger.warning("Failed to read data set {:d}.".format(i))
failed_list.append(nmolecules)
else:
res_list.append({
'nmolecules': np.array([nmolecules]*len(xvg_time), dtype=int),
'time': xvg_time,
**{i: xvg_data[:,i] for i in range(nmolecules)}
})
os.unlink(tmpin.name)
print('.',end='')
print('')
In [568]:
res_list
Out[568]:
In [581]:
res_df = None
for r in res_list:
#r['data'] = {i: r['data'][:,i] for i in range(r['data'].shape[1])}
cur_df = pd.DataFrame(r)
if res_df is None:
res_df = cur_df
else:
res_df = pd.merge(res_df, cur_df, how='outer', on=['nmolecules'])
res_df_mi = res_df.set_index(['nmolecules','time'])
In [583]:
res_df_mi
Out[583]:
In [590]:
res_df_mi.plot(legend=False)
Out[590]:
In [591]:
res_df_mi.mean(axis=1).plot()
Out[591]:
Pulling groups movement¶
In [733]:
res_list = []
failed_list = []
query = {
"metadata.project": project_id,
"metadata.type": 'pullx_xvg',
}
fp.filepad.count_documents(query)
match_aggregation = {
"$match": query
}
sort_aggregation = {
"$sort": {
"metadata.datetime": pymongo.DESCENDING,
}
}
group_aggregation = {
"$group": {
"_id": { p: '$metadata.{}'.format(p) for p in parameter_names },
"degeneracy": {"$sum": 1}, # number matching data sets
"latest": {"$first": "$gfs_id"} # unique gridfs id of file
}
}
aggregation_pipeline = [ match_aggregation, sort_aggregation, group_aggregation ]
cursor = fp.filepad.aggregate(aggregation_pipeline)
for i, c in enumerate(cursor):
content, metadata = fp.get_file_by_id(c["latest"])
nmolecules = int(metadata["metadata"]["nmolecules"])
tmpin = tempfile.NamedTemporaryFile(mode='w+b',suffix='.xvg', delete=False)
with tmpin:
tmpin.write(content)
try:
xvg = gromacs.fileformats.XVG(tmpin.name)
xvg_time = xvg.array[0,:]
#xvg_labels = ['1', '1 ref', '1 dX', '1 dY', '1 dZ', '1 g 1 X', '1 g 1 Y', '1 g 1 Z', '1 g 2 X', '1 g 2 Y', '1 g 2 Z']
N_pull_coords = nmolecules
N_cols = len(xvg.names)
N_cols_per_coord = int(N_cols / N_pull_coords)
xvg_labels = xvg.names[:N_cols_per_coord]
xvg_data = {}
for j in range(N_pull_coords):
for k in range(N_cols_per_coord):
xvg_data[(j,xvg_labels[k])] = xvg.array[
1+j*N_cols_per_coord+k,:]
except:
logger.exception("Failed to read data set {:d}.".format(i))
failed_list.append(nmolecules)
else:
res_list.append({
'nmolecules': nmolecules, # np.array([nmolecules]*len(xvg_time), dtype=int),
'time': xvg_time,
**xvg_data})
#os.unlink(tmpin.name)
print('.',end='')
print('')
In [734]:
res_list
Out[734]:
In [743]:
res_df = None
for r in res_list:
#r['data'] = {i: r['data'][:,i] for i in range(r['data'].shape[1])}
cur_df = pd.DataFrame(r)
if res_df is None:
res_df = cur_df
else:
res_df = pd.merge(res_df, cur_df, how='outer', on=['nmolecules'])
res_df_mi = res_df.set_index(['nmolecules','time'])
res_df_mi.columns = pd.MultiIndex.from_tuples(res_df_mi.columns, names=['nmolecule', 'coord'])
In [750]:
res_df_mi.groupby(axis=1,level='coord').mean()
Out[750]:
In [780]:
res_df_mi[0,'1'].plot()
Out[780]:
In [781]:
res_df_mi.groupby(axis=1,level='coord').mean()['1'].plot()
Out[781]:
In [782]:
res_df_mi.groupby(axis=1,level='coord').mean()['1 ref'].plot()
Out[782]:
In [472]:
# sqrt(dx^2+dy^2+dz^2), the distance between pulling groups (i.e. one surfactant tail atom and the Au COM)
for i in range(0,150,30):
plt.plot(pull_x_xvg.array[0,:],
np.sqrt(pull_x_xvg.array[i*11+3,:]**2+pull_x_xvg.array[i*11+4,:]**2+pull_x_xvg.array[i*11+5,:]**2))
plt.legend()
Out[472]:
In [473]:
pull_x_xvg.plot(columns=[0,12])
Out[473]:
Visualize trajectory¶
In [795]:
# Building a rather sophisticated aggregation pipeline
parameter_names = ['nmolecules', 'type']
query = {
"metadata.project": project_id,
"metadata.type": { '$in': ['pull_trr','pull_gro'] },
}
aggregation_pipeline = []
aggregation_pipeline.append({
"$match": query
})
aggregation_pipeline.append({
"$sort": {
"metadata.nmolecules": pymongo.ASCENDING,
"metadata.datetime": pymongo.DESCENDING,
}
})
aggregation_pipeline.append({
"$group": {
"_id": { p: '$metadata.{}'.format(p) for p in parameter_names },
"degeneracy": {"$sum": 1}, # number matching data sets
"latest": {"$first": "$gfs_id"} # unique gridfs id of file
}
})
parameter_names = ['nmolecules']
aggregation_pipeline.append({
"$group": {
"_id": { p: '$_id.{}'.format(p) for p in parameter_names },
"type": {"$addToSet": "$_id.type"},
"gfs_id": {"$addToSet": "$latest"}
#"$_id.type": "$latest"
}
})
aggregation_pipeline.append({
'$project': {
'_id': False,
**{ p: '$_id.{}'.format(p) for p in parameter_names},
'objects': {
'$zip': {
'inputs': [ '$type', '$gfs_id' ],
'useLongestLength': True,
'defaults': [None,None]
}
}
}
})
aggregation_pipeline.append({
'$project': {
**{ p: True for p in parameter_names},
'objects': {'$arrayToObject': '$objects'}
#'objects': False
}
})
aggregation_pipeline.append({
'$addFields': {
'objects': { **{ p: '${}'.format(p) for p in parameter_names} }
}
})
aggregation_pipeline.append({
'$replaceRoot': { 'newRoot': '$objects' }
})
# display results with
# for i, c in enumerate(cursor):
# print(c)
# yields documents in the form
# {'em_gro': '5e6a4e3d6c26f976ceae5e38', 'em_trr': '5e6a4e3a6c26f976ceae5e14', 'nmolecules': '44'}
# i.e. most recent topology file and trajectory file per concentration
cursor = fp.filepad.aggregate(aggregation_pipeline)
In [797]:
failed_list
mda_trr_list = []
for i, c in enumerate(cursor):
try:
gro_content, _ = fp.get_file_by_id(c["pull_gro"])
trr_content, _ = fp.get_file_by_id(c["pull_trr"])
with tempfile.NamedTemporaryFile(suffix='.gro') as gro, \
tempfile.NamedTemporaryFile(suffix='.trr') as trr:
gro.write(gro_content)
trr.write(trr_content)
mda_trr_list.append( mda.Universe(gro.name,trr.name) )
except:
logger.exception("Failed to read data set {}.".format(c))
failed_list.append(c)
print('.',end='')
print('')
In [798]:
failed_list
Out[798]:
In [799]:
mda_trr = mda_trr_list[0]
mda_view = nglview.show_mdanalysis(mda_trr)
mda_view.clear_representations()
mda_view.background = 'white'
mda_view.add_representation('ball+stick')
mda_view
MSD¶
In [476]:
substrate = mda_trr.atoms[mda_trr.atoms.names == 'AU']
In [477]:
surfactant_head = mda_trr.atoms[mda_trr.atoms.names == 'S']
In [478]:
rms_substrate = mda_rms.RMSD(substrate,ref_frame=0)
In [479]:
rms_substrate.run()
Out[479]:
In [480]:
rmsd = rms_substrate.rmsd.T # transpose makes it easier for plotting
time = rmsd[1]
In [481]:
plt.plot(time,rmsd[2])
Out[481]:
In [482]:
rms_surfactant_head = mda_rms.RMSD(surfactant_head,ref_frame=0)
In [483]:
rms_surfactant_head.run()
Out[483]:
In [484]:
rmsd = rms_surfactant_head.rmsd.T # transpose makes it easier for plotting
time = rmsd[1]
In [485]:
plt.plot(time,rmsd[2])
Out[485]:
Au-S (substrate - head group )RDF¶
In [486]:
len(mda_trr.trajectory)
Out[486]:
In [487]:
rdf_substrate_headgroup = mda_rdf.InterRDF(
substrate,surfactant_head,range=(0.0,80.0),verbose=True)
In [488]:
bins = []
rdf = []
for i in range(len(mda_trr.trajectory)):
rdf_substrate_headgroup = mda_rdf.InterRDF(
substrate,surfactant_head,range=(0.0,80.0),verbose=True)
rdf_substrate_headgroup.run(start=i,stop=i+1)
bins.append(rdf_substrate_headgroup.bins.copy())
rdf.append(rdf_substrate_headgroup.rdf.copy())
bins = np.array(bins)
rdf = np.array(rdf)
In [489]:
# indicates desired approach towards substrate
plt.plot(bins[0],rdf[0],label="Initial RDF")
plt.plot(bins[3],rdf[4],label="Intermediat RDF")
plt.plot(bins[-1],rdf[-1],label="Final RDF")
plt.legend()
Out[489]:
Single system global observables¶
In [457]:
edr_file = 'pull.edr'
In [458]:
edr_df = panedr.edr_to_df(edr_file)
In [459]:
edr_df.columns
Out[459]:
In [460]:
fig, ax = plt.subplots(3,2,figsize=(10,12))
edr_df.plot('Time','Potential',ax=ax[0,0])
edr_df.plot('Time','Pressure',ax=ax[0,1])
#edr_df.plot('Time','Bond',ax=ax[1,0])
edr_df.plot('Time','Position Rest.',ax=ax[1,0])
edr_df.plot('Time','COM Pull En.',ax=ax[1,1])
edr_df.plot('Time','Coulomb (SR)',ax=ax[2,0])
edr_df.plot('Time','Coul. recip.',ax=ax[2,1])
Out[460]:
Pulling forces¶
In [461]:
# read xvg file
pull_f_xvg = mda.auxiliary.XVG.XVGFileReader('pull_pullf.xvg')
pull_f_t = pull_f_xvg.read_all_times()
# first data column contains time, strip
pull_f = np.array([ f.data[1:] for f in pull_f_xvg ])
In [462]:
for i in range(0,199,50):
plt.plot(pull_f_t,pull_f[:,i])
Pulling groups movement¶
In [463]:
pull_x_xvg = gromacs.fileformats.XVG('pull_pullx.xvg',)
In [464]:
pull_x_xvg.array
Out[464]:
In [465]:
len(pull_x_xvg.names)
Out[465]:
In [466]:
# that many columns perr pull coordinate
N_cols_per_coord = int(len(pull_x_xvg.names) / N_pull_coords)
In [467]:
# with content
legend = pull_x_xvg.names[:11]
In [468]:
legend
Out[468]:
In [469]:
pull_x_xvg.names[-3:]
Out[469]:
In [470]:
for i in range(11):
plt.plot(pull_x_xvg.array[0,:],pull_x_xvg.array[i+1,:],label=legend[i])
plt.legend()
Out[470]:
In [471]:
for i in range(11):
plt.plot(pull_x_xvg.array[0,:],pull_x_xvg.array[i+1,:],label=legend[i])
plt.legend()
Out[471]:
In [472]:
# sqrt(dx^2+dy^2+dz^2), the distance between pulling groups (i.e. one surfactant tail atom and the Au COM)
for i in range(0,150,30):
plt.plot(pull_x_xvg.array[0,:],
np.sqrt(pull_x_xvg.array[i*11+3,:]**2+pull_x_xvg.array[i*11+4,:]**2+pull_x_xvg.array[i*11+5,:]**2))
plt.legend()
Out[472]:
In [473]:
pull_x_xvg.plot(columns=[0,12])
Out[473]:
Visualize trajectory¶
In [272]:
gro_em = 'pull.gro'
In [273]:
mda_trr = mda.Universe('em.gro','pull.trr')
mda_view = nglview.show_mdanalysis(mda_trr)
mda_view.clear_representations()
mda_view.background = 'white'
mda_view.add_representation('ball+stick')
mda_view
In [536]:
mda_xtc = mda.Universe(gro,'pull.xtc')
mda_view = nglview.show_mdanalysis(mda_xtc)
mda_view.clear_representations()
mda_view.background = 'white'
mda_view.add_representation('ball+stick')
mda_view
MSD¶
In [476]:
substrate = mda_trr.atoms[mda_trr.atoms.names == 'AU']
In [477]:
surfactant_head = mda_trr.atoms[mda_trr.atoms.names == 'S']
In [478]:
rms_substrate = mda_rms.RMSD(substrate,ref_frame=0)
In [479]:
rms_substrate.run()
Out[479]:
In [480]:
rmsd = rms_substrate.rmsd.T # transpose makes it easier for plotting
time = rmsd[1]
In [481]:
plt.plot(time,rmsd[2])
Out[481]:
In [482]:
rms_surfactant_head = mda_rms.RMSD(surfactant_head,ref_frame=0)
In [483]:
rms_surfactant_head.run()
Out[483]:
In [484]:
rmsd = rms_surfactant_head.rmsd.T # transpose makes it easier for plotting
time = rmsd[1]
In [485]:
plt.plot(time,rmsd[2])
Out[485]:
Au-S (substrate - head group )RDF¶
In [486]:
len(mda_trr.trajectory)
Out[486]:
In [487]:
rdf_substrate_headgroup = mda_rdf.InterRDF(
substrate,surfactant_head,range=(0.0,80.0),verbose=True)
In [488]:
bins = []
rdf = []
for i in range(len(mda_trr.trajectory)):
rdf_substrate_headgroup = mda_rdf.InterRDF(
substrate,surfactant_head,range=(0.0,80.0),verbose=True)
rdf_substrate_headgroup.run(start=i,stop=i+1)
bins.append(rdf_substrate_headgroup.bins.copy())
rdf.append(rdf_substrate_headgroup.rdf.copy())
bins = np.array(bins)
rdf = np.array(rdf)
In [489]:
# indicates desired approach towards substrate
plt.plot(bins[0],rdf[0],label="Initial RDF")
plt.plot(bins[3],rdf[4],label="Intermediat RDF")
plt.plot(bins[-1],rdf[-1],label="Final RDF")
plt.legend()
Out[489]:
Solvation¶
Now, fill the box with water.
In [560]:
gro = 'pull.gro'
In [561]:
# use -scale 0.5 -maxsol N for non-standard conditions
gmx_solvate = gromacs.solvate.Popen(
cp=gro, cs='spc216.gro',o='solvated.gro',p=top,
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
out = gmx_solvate.stdout.read()
err = gmx_solvate.stderr.read()
In [562]:
print(out)
In [563]:
print(err)
Energy minimization with restraints¶
Again, relax the system a little with positional constraints applied to all ions.
Execute trial task via Fireworks on remote resource¶
In [20]:
lpad = LaunchPad.auto_load()
A trial task sent to FORHLR2:
In [71]:
gmx_test_task = CmdTask(
cmd = 'gmx',
opt = '-h',
stderr_file = 'std.err',
stdout_file = 'std.out',
use_shell = True)
In [72]:
gmx_test_fw = Firework(
gmx_test_task,
name = 'FORHLR2 GMX test fw',
spec = {
'_category': 'forhlr2_noqueue',
'_files_out': {
'stdout': 'std.out',
'stderr': 'std.err'}
} )
In [73]:
fw_ids = lpad.add_wf(gmx_test_fw)
In [74]:
fw_ids
Out[74]:
In [86]:
# lpad.delete_wf(INSERT_ID,delete_launch_dirs=True)
Compile system¶
In [237]:
top = 'sys.top'
In [238]:
gro = 'solvated.gro'
In [137]:
gmx_grompp = gromacs.grompp.Popen(
f='em_solvated.mdp',c=gro,r=gro,o='em_solvated.tpr',p=top,
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
out = gmx_grompp.stdout.read()
err = gmx_grompp.stderr.read()
In [138]:
print(err)
In [139]:
print(out)
Remote file transfer¶
Utilize fabric to transfer files files to remote resource conveniently:
In [30]:
c = fabric.Connection('forhlr2') # host defined in ssh config
In [31]:
res = c.run('ws_find fw') # get remote directory of Fireworks workspace
In [32]:
res.command
Out[32]:
In [144]:
now = datetime.now().isoformat()
In [145]:
remote_path = os.path.sep.join((res.stdout.strip(),'file_transfer',now))
In [194]:
remote_path
Out[194]:
In [147]:
res = c.run(' '.join(['mkdir','-p',remote_path]))
In [148]:
file_name = 'em_solvated.tpr'
In [188]:
local_file = os.path.sep.join((prefix,file_name))
In [189]:
remote_file = os.path.sep.join((remote_path,file_name))
In [198]:
res = c.put(local_file,remote_file)
In [200]:
res.local
Out[200]:
In [193]:
# FileTransferTask does not work anymore
In [191]:
#ft = FileTransferTask(
# mode = 'rtransfer',
# files = [ {'src':local_file, 'dest':remote_path} ],
# server = c.host,
# user = c.user )
In [192]:
#fw = Firework(
# ft,
# name = 'BWCLOUD remote transef to FORHLR2',
# spec = {
# '_category': 'bwcloud_std',
# } )
In [174]:
fw_ids = lpad.add_wf(fw)
Run energy minimization¶
In [239]:
ft = FileTransferTask(
mode = 'copy',
files = [ {'src':remote_file, 'dest':os.path.curdir} ] )
In [240]:
gmx_mdrun_task = CmdTask(
cmd = 'gmx',
opt = ['mdrun','-v','-deffnm','em_solvated'],
stderr_file = 'std.err',
stdout_file = 'std.out',
use_shell = True)
In [241]:
gmx_log_tracker = Tracker('em_solvated.log')
In [242]:
gmx_mdrun_fw = Firework(
[ft,gmx_mdrun_task],
name = 'FORHLR2 GMX mdrun em_solvated',
spec = {
'_category': 'forhlr2_queue',
'_queueadapter': {
'cpus_per_task': 1,
'ntasks_per_node': 20,
'ntasks': 40,
'queue': 'normal',
'walltime': '24:00'
},
'_files_out': {
'log': '*.log',
'trr': '*.trr',
'edr': '*.edr',
'gro': '*.gro' },
'_trackers' : [ gmx_log_tracker ]
} )
In [243]:
pprint(gmx_mdrun_fw.as_dict())
In [244]:
fw_ids = lpad.add_wf(gmx_mdrun_fw)
In [315]:
fw_id = list(fw_ids.values())[0]
File transfer back¶
instead of relying on the returned fw_id, we can also query the Firework added latest
In [23]:
fw_ids = lpad.get_fw_ids(sort=[('created_on',pymongo.DESCENDING)],limit=1)
In [25]:
fw_id = fw_ids[0]
In [34]:
fw_id
Out[34]:
We query the remote directory our FireWork ran in
In [28]:
launch_dir = lpad.get_launchdir(fw_id)
In [35]:
launch_dir
Out[35]:
In [30]:
c = fabric.Connection('forhlr2') # host defined in ssh config
In [36]:
res = c.run('ls -lht {}'.format(launch_dir)) # look at remote directory contents
In [39]:
glob_pattern = os.path.join(launch_dir,'em_solvated.*')
In [49]:
res = c.run('ls {}'.format(glob_pattern))
In [53]:
res.stdout
Out[53]:
In [57]:
for f in res.stdout.splitlines():
c.get(f)
Energy minimization analysis¶
In [97]:
em_file = 'em_solvated.edr'
In [98]:
em_df = panedr.edr_to_df(em_file)
In [99]:
em_df.columns
Out[99]:
In [100]:
fig, ax = plt.subplots(3,2,figsize=(10,12))
em_df.plot('Time','Potential',ax=ax[0,0])
em_df.plot('Time','Pressure',ax=ax[0,1])
em_df.plot('Time','Bond',ax=ax[1,0])
em_df.plot('Time','Position Rest.',ax=ax[1,1])
#em_df.plot('Time','COM Pull En.',ax=ax[1,1])
em_df.plot('Time','Coulomb (SR)',ax=ax[2,0])
em_df.plot('Time','Coul. recip.',ax=ax[2,1])
Out[100]:
In [173]:
try:
del em_df
except:
pass
Visualize trajectory¶
In [102]:
mda_trr = mda.Universe('solvated.gro','em_solvated.trr')
In [103]:
# check unique resiude names in system
resnames = np.unique([ r.resname for r in mda_trr.residues ])
In [104]:
resnames
Out[104]:
In [117]:
mda_view = nglview.show_mdanalysis(mda_trr)
mda_view.clear_representations()
mda_view.background = 'white'
mda_view.add_representation(repr_type='ball+stick',selection='SDS')
mda_view.add_representation(repr_type='ball+stick',selection='NA')
mda_view.add_representation(repr_type='spacefill',selection='AUM',color='yellow')
mda_view.center()
In [118]:
mda_view
In [159]:
try:
del mda_trr
except:
pass
try:
del mda_view
except:
pass
NVT equilibration¶
In [134]:
top = 'sys.top'
gro = 'em_solvated.gro'
ndx = 'nvt.ndx'
In [133]:
lpad = LaunchPad.auto_load()
Generate non-substrate index group¶
In [141]:
pmd_top_gro = pmd.gromacs.GromacsTopologyFile(top)
pmd_gro = pmd.gromacs.GromacsGroFile.parse(gro)
pmd_top_gro.box = pmd_gro.box
pmd_top_gro.positions = pmd_gro.positions
In [142]:
non_substrate_ndx = np.array([
i+1 for i,a in enumerate(pmd_top_gro.atoms) if a.residue.name != 'AUM' ])
# gromacs ndx starts at 1
len(pmd_top_gro.atoms)
In [144]:
len(non_substrate_ndx)
Out[144]:
In [147]:
len(pmd_top_gro.atoms) - len(non_substrate_ndx) # double-check non-substrate and substrate atom numbers
Out[147]:
In [160]:
try:
del pmd_top_gro
except:
pass
try:
del pmd_gro
except:
pass
Generate standard index file for system¶
In [135]:
gmx_make_ndx = gromacs.make_ndx.Popen(
f=gro,o=ndx,
input='q',
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
In [136]:
out_str, err_str = gmx_make_ndx.communicate()
In [137]:
print(out_str)
In [138]:
print(err_str)
Enhance standard index file by pulling groups¶
In [150]:
ndx_in = gromacs.fileformats.NDX(ndx)
In [152]:
ndx_in['non-Substrate'] = non_substrate_ndx
In [155]:
ndx_in.write(ndx)
In [175]:
try:
del ndx_in
except:
pass
try:
del non_substrate_ndx
except:
pass
Compile system¶
In [183]:
gmx_grompp = gromacs.grompp.Popen(
f='nvt.mdp',n=ndx,c=gro,r=gro,o='nvt.tpr',p=top,
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
out = gmx_grompp.stdout.read()
err = gmx_grompp.stderr.read()
In [184]:
print(err)
In [185]:
print(out)
Remote file transfer¶
Utilize fabric to transfer files files to remote resource conveniently:
In [279]:
c = fabric.Connection('forhlr2') # host defined in ssh config
In [280]:
res = c.run('ws_find fw') # get remote directory of Fireworks workspace
In [281]:
res.command
Out[281]:
In [282]:
now = datetime.now().isoformat()
In [283]:
remote_path = os.path.sep.join((res.stdout.strip(),'file_transfer',now))
In [191]:
remote_path
Out[191]:
In [192]:
res = c.run(' '.join(['mkdir','-p',remote_path]))
In [193]:
file_name = 'nvt.tpr'
In [194]:
local_file = os.path.sep.join((prefix,file_name))
In [195]:
remote_file = os.path.sep.join((remote_path,file_name))
In [196]:
res = c.put(local_file,remote_file)
In [197]:
res.local
Out[197]:
Execute trial task via Fireworks on remote resource queue¶
In [317]:
ft = FileTransferTask(
mode = 'copy',
files = [ {'src':remote_file, 'dest':os.path.curdir} ] )
In [318]:
gmx_mdrun_task = CmdTask(
cmd = 'gmx',
opt = ['mdrun','-v','-deffnm','nvt'],
stderr_file = 'std.err',
stdout_file = 'std.out',
use_shell = True)
In [319]:
gmx_log_tracker = Tracker('nvt.log')
In [320]:
gmx_mdrun_fw = Firework(
[ft,gmx_mdrun_task],
name = 'FORHLR2 GMX mdrun nvt',
spec = {
'_category': 'forhlr2_queue',
'_queueadapter': {
'cpus_per_task': 1,
'ntasks_per_node': 20,
'ntasks': 20,
'queue': 'develop',
'walltime': '00:03:00'
},
'_files_out': {
'log': '*.log',
'trr': '*.trr',
'edr': '*.edr',
'gro': '*.gro' },
'_trackers' : [ gmx_log_tracker ]
} )
In [321]:
pprint(gmx_mdrun_fw.as_dict())
In [322]:
fw_ids = lpad.add_wf(gmx_mdrun_fw)
In [332]:
fw_ids = lpad.get_fw_ids(query={'name':'FORHLR2 GMX mdrun nvt','spec._queueadapter.queue':'develop'})
In [333]:
for fw_id in fw_ids:
try:
print("Deleting {}...".format(fw_id))
lpad.delete_wf(fw_id)
except:
print("Failed deleting {}...".format(fw_id))
Run NVT equilibration¶
In [198]:
ft = FileTransferTask(
mode = 'copy',
files = [ {'src':remote_file, 'dest':os.path.curdir} ] )
In [199]:
gmx_mdrun_task = CmdTask(
cmd = 'gmx',
opt = ['mdrun','-v','-deffnm','nvt'],
stderr_file = 'std.err',
stdout_file = 'std.out',
use_shell = True)
In [200]:
gmx_log_tracker = Tracker('nvt.log')
In [201]:
gmx_mdrun_fw = Firework(
[ft,gmx_mdrun_task],
name = 'FORHLR2 GMX mdrun nvt',
spec = {
'_category': 'forhlr2_queue',
'_queueadapter': {
'cpus_per_task': 1,
'ntasks_per_node': 20,
'ntasks': 40,
'queue': 'normal',
'walltime': '24:00'
},
'_files_out': {
'log': '*.log',
'trr': '*.trr',
'edr': '*.edr',
'gro': '*.gro' },
'_trackers' : [ gmx_log_tracker ]
} )
In [202]:
pprint(gmx_mdrun_fw.as_dict())
In [203]:
fw_ids = lpad.add_wf(gmx_mdrun_fw)
In [204]:
fw_id = list(fw_ids.values())[0]
File transfer back¶
instead of relying on the returned fw_id, we can also query the Firework added latest
In [205]:
fw_ids = lpad.get_fw_ids(sort=[('created_on',pymongo.DESCENDING)],limit=1)
In [206]:
fw_id = fw_ids[0]
In [207]:
fw_id
Out[207]:
We query the remote directory our FireWork ran in
In [208]:
launch_dir = lpad.get_launchdir(fw_id)
In [209]:
launch_dir
Out[209]:
In [210]:
c = fabric.Connection('forhlr2') # host defined in ssh config
In [211]:
res = c.run('ls -lht {}'.format(launch_dir)) # look at remote directory contents
In [213]:
glob_pattern = os.path.join(launch_dir,'nvt.*')
In [214]:
res = c.run('ls {}'.format(glob_pattern))
In [215]:
for f in res.stdout.splitlines():
c.get(f)
Analysis¶
In [242]:
edr_file = 'nvt.edr'
In [243]:
edr_df = panedr.edr_to_df(edr_file)
In [244]:
edr_df.columns
Out[244]:
In [248]:
fig, ax = plt.subplots(3,2,figsize=(10,12))
edr_df.plot('Time','Temperature',ax=ax[0,0])
edr_df.plot('Time','Pressure',ax=ax[0,1])
edr_df.plot('Time','Potential',ax=ax[1,0])
edr_df.plot('Time','Bond',ax=ax[1,1])
#edr_df.plot('Time','Position Rest.',ax=ax[1,1])
#edr_df.plot('Time','COM Pull En.',ax=ax[1,1])
edr_df.plot('Time','Coulomb (SR)',ax=ax[2,0])
edr_df.plot('Time','Coul. recip.',ax=ax[2,1])
Out[248]:
Visualize trajectory¶
In [102]:
mda_trr = mda.Universe('solvated.gro','em_solvated.trr')
In [103]:
# check unique resiude names in system
resnames = np.unique([ r.resname for r in mda_trr.residues ])
In [104]:
resnames
Out[104]:
In [117]:
mda_view = nglview.show_mdanalysis(mda_trr)
mda_view.clear_representations()
mda_view.background = 'white'
mda_view.add_representation(repr_type='ball+stick',selection='SDS')
mda_view.add_representation(repr_type='ball+stick',selection='NA')
mda_view.add_representation(repr_type='spacefill',selection='AUM',color='yellow')
mda_view.center()
In [118]:
mda_view
In [119]:
try:
del mda_trr
except:
pass
try:
del mda_view
except:
pass
NPT equilibration¶
In [110]:
top = 'sys.top'
gro = 'nvt.gro'
ndx = 'nvt.ndx'
In [111]:
lpad = LaunchPad.auto_load()
Compile system¶
In [117]:
gmx_grompp = gromacs.grompp.Popen(
f='npt.mdp',n=ndx,c=gro,r=gro,o='npt.tpr',p=top,
stdout=subprocess.PIPE,stderr=subprocess.PIPE)
out = gmx_grompp.stdout.read()
err = gmx_grompp.stderr.read()
In [118]:
print(err)
In [119]:
print(out)
Remote file transfer¶
Utilize fabric to transfer files files to remote resource conveniently:
In [120]:
c = fabric.Connection('forhlr2') # host defined in ssh config
In [121]:
res = c.run('ws_find fw') # get remote directory of Fireworks workspace
In [122]:
res.command
Out[122]:
In [123]:
now = datetime.now().isoformat()
In [124]:
remote_path = os.path.sep.join((res.stdout.strip(),'file_transfer',now))
In [125]:
remote_path
Out[125]:
In [126]:
res = c.run(' '.join(['mkdir','-p',remote_path]))
In [127]:
file_name = 'npt.tpr'
In [128]:
local_file = os.path.sep.join((prefix,file_name))
In [129]:
remote_file = os.path.sep.join((remote_path,file_name))
In [130]:
res = c.put(local_file,remote_file)
Run NPT equilibration¶
In [131]:
ft = FileTransferTask(
mode = 'copy',
files = [ {'src':remote_file, 'dest':os.path.curdir} ] )
In [132]:
gmx_mdrun_task = CmdTask(
cmd = 'gmx',
opt = ['mdrun','-v','-deffnm','npt'],
stderr_file = 'std.err',
stdout_file = 'std.out',
use_shell = True)
In [133]:
gmx_log_tracker = Tracker('npt.log')
In [134]:
gmx_mdrun_fw = Firework(
[ft,gmx_mdrun_task],
name = 'FORHLR2 GMX mdrun npt',
spec = {
'_category': 'forhlr2_queue',
'_queueadapter': {
'cpus_per_task': 1,
'ntasks_per_node': 20,
'ntasks': 20,
'queue': 'normal',
'walltime': '24:00:00'
},
'_files_out': {
'log': '*.log',
'trr': '*.trr',
'edr': '*.edr',
'gro': '*.gro' },
'_trackers' : [ gmx_log_tracker ]
} )
In [135]:
fw_ids = lpad.add_wf(gmx_mdrun_fw)
In [136]:
fw_id = list(fw_ids.values())[0]
In [137]:
fw_id
Out[137]:
File transfer back¶
instead of relying on the returned fw_id, we can also query the Firework added latest
In [47]:
fw_ids = lpad.get_fw_ids(sort=[('created_on',pymongo.DESCENDING)],limit=1)
In [48]:
fw_id = fw_ids[0]
In [49]:
fw_id
Out[49]:
We query the remote directory our FireWork ran in
In [50]:
launch_dir = lpad.get_launchdir(fw_id)
In [51]:
launch_dir
Out[51]:
In [52]:
c = fabric.Connection('forhlr2') # host defined in ssh config
In [53]:
res = c.run('ls -lht {}'.format(launch_dir)) # look at remote directory contents
In [54]:
glob_pattern = os.path.join(launch_dir,'npt.*')
In [55]:
res = c.run('ls {}'.format(glob_pattern))
In [56]:
for f in res.stdout.splitlines():
c.get(f)
Analysis¶
In [57]:
edr_file = 'npt.edr'
In [58]:
edr_df = panedr.edr_to_df(edr_file)
In [59]:
edr_df.columns
Out[59]:
In [60]:
fig, ax = plt.subplots(3,2,figsize=(10,12))
edr_df.plot('Time','Temperature',ax=ax[0,0])
edr_df.plot('Time','Pressure',ax=ax[0,1])
edr_df.plot('Time','Potential',ax=ax[1,0])
edr_df.plot('Time','Bond',ax=ax[1,1])
#edr_df.plot('Time','Position Rest.',ax=ax[1,1])
#edr_df.plot('Time','COM Pull En.',ax=ax[1,1])
edr_df.plot('Time','Coulomb (SR)',ax=ax[2,0])
edr_df.plot('Time','Coul. recip.',ax=ax[2,1])
Out[60]:
Visualize trajectory¶
In [61]:
mda_trr = mda.Universe('nvt.gro','npt.trr')
In [62]:
# check unique resiude names in system
resnames = np.unique([ r.resname for r in mda_trr.residues ])
In [63]:
resnames
Out[63]:
In [64]:
mda_view = nglview.show_mdanalysis(mda_trr)
mda_view.clear_representations()
mda_view.background = 'white'
mda_view.add_representation(repr_type='ball+stick',selection='SDS')
mda_view.add_representation(repr_type='ball+stick',selection='NA')
mda_view.add_representation(repr_type='spacefill',selection='AUM',color='yellow')
mda_view.center()
In [65]:
mda_view
In [67]:
substrate = mda_trr.select_atoms('resname AUM')
In [103]:
substrate.masses= ase.data.atomic_masses[ase.data.atomic_numbers['Au']]
In [105]:
substrtate_com_traj = np.array([substrate.center_of_mass() for ts in mda_trr.trajectory ])
In [106]:
substrtate_rgyr_traj = np.array([substrate.radius_of_gyration() for ts in mda_trr.trajectory ])
In [107]:
fig = plt.figure(figsize=(12,10))
ax = fig.add_subplot(111, projection='3d',azim=-30)
ax.plot(*substrtate_com_traj.T)
ax.scatter(*substrtate_com_traj[0,:],color='green')
ax.scatter(*substrtate_com_traj[-1,:],color='red')
Out[107]:
In [109]:
plt.plot(substrtate_rgyr_traj)
Out[109]:
In [119]:
try:
del mda_trr
except:
pass
try:
del mda_view
except:
pass